Technology

DeepSeek landed like a thunderclap in January 2025, when its R1 reasoning model briefly dethroned ChatGPT as the most downloaded free app on the iOS App Store in the United States. Built by a Hangzhou-based AI lab backed by Chinese hedge fund High-Flyer, it claimed to match frontier AI performance at a fraction of the development cost. That claim sent Nvidia's stock tumbling 17% in a single session and sparked a global conversation about who was actually winning the AI race.

Since then, DeepSeek has grown to roughly 97 million monthly active users and released multiple model generations, most recently the V4 family in April 2026. Its API pricing stands out: the V4 Flash model starts at $0.14 per million input tokens, cheaper than most "lite" tier models from OpenAI and Google yet competitive on coding, math, and reasoning benchmarks. The open-weight licensing under MIT also means teams can self-host the models and sidestep per-token costs entirely at scale.

We've been reviewing B2B software at TechRadar Pro since 2012, with AI platforms among our most active coverage areas in recent years. Our AI tools roundup, vibe coding guide for 2026, and explainers on OpenClaw and Moltbook give you a sense of the tools we track. DeepSeek is one of the more polarizing platforms we've tested: impressive in many ways, but not without significant red flags.

What is DeepSeek?

DeepSeek is an AI chat platform and API service developed by Hangzhou DeepSeek Artificial Intelligence Co., Ltd., a Chinese company founded in 2023 and funded by the quant hedge fund High-Flyer. It offers a free web and mobile chat interface at chat.deepseek.com alongside a paid developer API, both powered by the same underlying model family.

The platform runs on DeepSeek's own large language models, specifically V4 Flash and V4 Pro, both using a Mixture-of-Experts (MoE) architecture. Only a subset of each model's parameters activates per token, which keeps inference costs low without shrinking the model's overall knowledge base.

V4 Pro has 1.6 trillion total parameters but only 49 billion active at inference. V4 Flash runs 284 billion total with 13 billion active, making it significantly faster and cheaper without sacrificing much on everyday tasks.

Developers, researchers, and cost-conscious businesses are the natural audience. The free chat tier suits individuals and small teams exploring the tool, while the API's aggressive pricing makes it attractive for anyone building AI-powered applications at scale.

DeepSeek: At a glance

Attribute

Notes

Underlying model(s)

DeepSeek V4 Flash (284B total / 13B active params) and V4 Pro (1.6T total / 49B active params), both MoE-based

Best for

Coding assistance, mathematical reasoning, document analysis, budget API use

Distinguishing functions

1M token context, thinking/non-thinking modes, prompt caching, open weights (MIT)

UI features

Web chat and iOS/Android apps with web search toggle, file upload (PDF, DOCX, TXT), Expert Mode and Instant Mode

Subscription costs

Free (chat app, unlimited queries); no paid chat subscription tiers

API pricing

Pay-per-token; new accounts receive 5M free tokens valid 30 days; V4 Flash at $0.14 / $0.28 per 1M tokens (input/output); V4 Pro at $1.74 / $3.48 standard, with promotional discounts available

Buy it if…

• You need a cheap, capable coding or reasoning API. V4 Flash at $0.14 per million input tokens is among the most affordable frontier-adjacent APIs available, and its benchmark results on coding and math hold up well against more expensive competitors.
• You want open weights for self-hosted deployment. Both V4 models are released under MIT license on Hugging Face, giving teams the option to run the model on their own infrastructure and eliminate per-token costs entirely for high-volume workloads.
• You're doing non-sensitive exploratory work. For individual researchers, students, or developers prototyping non-confidential projects, the free chat app offers web search, file uploads, and a 1M token context window at zero cost.

Don't buy it if…

• Your work involves confidential or regulated data. All user data is stored on servers in mainland China, subject to Chinese law, which permits government access without user consent. Multiple governments have banned DeepSeek from official devices for exactly this reason.
• You need consistent responses on sensitive topics. DeepSeek avoids certain politically sensitive subjects, particularly around Chinese domestic affairs. That content filtering can produce evasive or incomplete outputs on topics that other platforms handle straightforwardly.
• You're in a GDPR-regulated region. Italy's data protection authority blocked DeepSeek outright in January 2025 after the company provided what regulators called a "completely insufficient" response to data practice inquiries.

My time with DeepSeek

I tested DeepSeek's chat app and API across a range of tasks: code generation, document summarization, long-form reasoning, and general Q&A. On raw capability, the V4 models impressed me. Code outputs were clean and well-structured, long document summaries were accurate, and the one-million-token context window handled full-length PDF ingestion without complaint.

The thinking mode, accessible via Expert Mode in the chat UI, added visible chain-of-thought reasoning that proved useful for multi-step problems rather than theatrical.

What gave me pause was everything outside the model itself. Certain politically sensitive prompts returned conspicuously vague or deflective answers — the kind of behavior that wouldn't be acceptable in a professional context where consistent and complete information matters. I also found that the chat interface lacks the memory and personalization features you'd find in ChatGPT or Claude.

Value for money on the API side is difficult to argue with. A production app with well-structured prompts benefits substantially from the caching discount: cached input tokens cost just $0.014 per million for V4 Flash, a 90% reduction. For high-volume, low-sensitivity workloads, that arithmetic is compelling.

DeepSeek: Features

DeepSeek's core chat feature set covers the bases you'd expect: text generation, code writing and debugging, document summarization, mathematical reasoning, and web search. The web search integration is a manual toggle rather than always-on, which keeps responses faster by default but requires you to switch it on when real-time information matters. File uploads support PDF, DOCX, and TXT formats, with the model able to summarize and answer questions based on the uploaded content.

The standout capability is the 1M token context window introduced with V4, up from 128K in the previous generation. That's a meaningful jump for anyone analyzing long contracts, codebases, or research documents in a single session. Most competitors at comparable price points max out at 128K to 200K tokens.

V4 Flash covers both thinking and non-thinking modes, so you don't need to switch between separate models depending on task complexity. Non-thinking handles fast general responses; thinking adds structured multi-step reasoning for harder problems. That flexibility matters more than it sounds when you're toggling between casual tasks and complex analysis in the same workflow.

Where DeepSeek falls short is multimodal support. The platform does not currently support image generation or image understanding in the web app, putting it behind ChatGPT, Claude, and Gemini on that front. Agentic capabilities are available in the V4 Preview but remain early-stage compared to dedicated agentic platforms.

DeepSeek: User experience

The chat interface at chat.deepseek.com is straightforward and fast to get started with. Signing up requires only an email address from a global provider like Gmail or Yahoo, and the default experience drops you straight into a conversation window. The distinction between Expert Mode (thinking-enabled, slower) and Instant Mode (faster, non-thinking) is surfaced clearly at the top of the interface, and mobile apps on iOS and Android mirror the web experience with file upload and web search included.

The learning curve is shallow for casual use. Switching between thinking and non-thinking modes takes one click, and the file upload workflow is drag-and-drop simple.

The API experience is less forgiving for first-time integrators. Unlike the chat app, the API is stateless, meaning every call must include the full conversation history in the messages array. DeepSeek's documentation covers this clearly, but it catches developers accustomed to managed conversation state elsewhere off guard.

DeepSeek: Customer support

Support options for free chat users are limited to a Discord community server and an email channel for API service inquiries (api-service@deepseek.com). Community responses on Discord can be prompt, but they depend on other users rather than official staff. There is no live chat or phone support.

API customers have slightly more recourse through direct email support, though response times vary. The official documentation at api-docs.deepseek.com is thorough and well-organized, covering model details, pricing, rate limits, and code examples in both Python and curl. For developers comfortable with self-service documentation, it's adequate.

(Image credit: DeepSeek)DeepSeek: Pricing

• Chat app is free with no query limits. The web and mobile apps give you unlimited access to V4 Flash and V4 Pro at no cost, including web search, file uploads, and the full context window.
• New API accounts receive 5M free tokens, valid for 30 days, giving developers a zero-cost window to prototype and test.
• API billing is pay-as-you-go. V4 Flash costs $0.14 per million input tokens and $0.28 per million output tokens. V4 Pro runs $1.74 and $3.48 respectively at standard rates, with promotional discounts periodically dropping those to $0.435 and $0.87. Cached input tokens cost one-tenth of the standard input rate on both models.

The free chat tier is generous by any measure. Unlimited queries with a 1M context window puts it ahead of most free-tier competitors in raw access terms, and there's no paid chat subscription to worry about. Power users who need more control either stick with the free app or pay per token via the API.

On the API side, DeepSeek makes a strong case for developers managing costs at scale. Off-peak pricing discounts of up to 75% are available during 16:30–00:30 UTC, giving teams with flexible scheduling another cost lever. For production apps with well-structured prompts sharing a common system context, effective input costs can drop well below $0.02 per million tokens with caching applied.

DeepSeek: alternatives you should consider

• ChatGPT (OpenAI): The most capable all-around AI platform, with image understanding, voice, and memory. API pricing is higher, but US-based data residency and enterprise data agreements make it a safer choice for sensitive business workloads.
• Claude (Anthropic): Particularly strong for long-form writing and document analysis, with comparable context windows and clearer data handling policies. Claude Pro starts at $20/month for individual users.
• Gemini (Google): Tightly integrated with Google Workspace and built for multimodal work across text, image, and video. Gemini 2.5 Flash offers competitive API pricing with no Chinese data jurisdiction concerns.

How I tested DeepSeek

• Used the free web app for code generation, document summarization, mathematical reasoning, multi-turn Q&A, and file analysis, covering both Instant Mode and Expert Mode across each task type.
• Ran API calls across varying prompt sizes to verify context window behavior, caching discounts, and response consistency, following DeepSeek's own temperature guidance for different task types.
• Cross-referenced DeepSeek's official privacy policy, third-party security research from NowSecure and SecurityScorecard, and regulatory actions from Italy, Australia, South Korea, and US government bodies to build a complete picture of the data risk profile.

Beyond hands-on testing, I reviewed DeepSeek's official API documentation, the V4 technical report published on Hugging Face, and benchmark data from the April 2026 release. Pricing figures were sourced directly from the official DeepSeek API documentation and corroborated against third-party tracking services.

• Microsoft produced 20.3 million tons of CO2e emissions last year, up from 16.2 million tons
• Of the 37.5 million MWh of energy used, only 422,000 MWh came from unsustainable sources
• A rise in diesel/crude oil consumption and Scope 2 emissions are key drivers

In its latest sustainability report, Microsoft has admitted its greenhouse gas emissions actually rose 25.1% year-over-year from 16.2 million tons to 20.3 million tons in 2025.

With the company targeting a 2030 carbon negative deadline, rising emissions presents a major challenge that it must overcome, however current trends point to emissions continue to rise even further.

Microsoft said its rapid expansion of AI and cloud data centers as a key driver for rising emissions, and with more projects in the pipeline, this could be an ongoing challenge for years to come.

Microsoft's emissions are moving in the wrong direction

The company also noted its decision to stop buying short-term renewable energy certificates that do not directly support additional clean capacity. While the previous year's 16.2-million-ton figure was lower than last year's, it was largely offset by carbon credits and doesn't accurately represent the true emissions.

More broadly, Scope 2 and Scope 3 emissions are also under pressure from the continued data center expansion, due to electricity purchases, unsustainable construction materials and compute hardware. For example, Scope 2 emissions went from accounting for 1.6% of total emissions in FY24 to a staggering 13.3% in FY25.

As for fossil fuel use, the company saw a 51% rise in diesel and crude oil consumption despite reductions in natural gas (-6.5%), propane/LPG/jet fuel (-10%) and gasoline (-16%) use. Still, of the nearly 37.5 million MWh of energy the company used in FY25, only around 422,000 MWh came from non-renewable sources (per a separate data sheet).

However, despite expansion-related challenges, Microsoft did make significant progress to reducing overall emissions, hitting around 20 million tons last year instead of the 34 million tones it could have hit without work on carbon-free electricity, sustainable fuels, energy efficiency improvements and other supple chain refinements.

Microsoft isn't the only company battling the impacts of AI – Amazon also recently noted a 16% annual increase in emissions, while also blaming AI and data centers. Google also saw a 25% rise in emissions for its most recent full year.

IBM’s launch of its AI coding assistant “Bob” points to a much bigger shift in enterprise modernization. Across the industry, AI tools are being positioned as a way to make legacy systems easier to understand, assess and eventually modernize. And there is real value there.

Some of these tools can read thousands of lines of legacy code, identify deprecated APIs, summarize business logic and surface technical debt in minutes. For organizations carrying decades of operational history, that kind of visibility is a big step forward - but let’s not confuse visibility with modernization.

Understanding how a system works is necessary. It is not sufficient. I have seen teams produce clean dependency maps, detailed code summaries and impressive technical assessments, only to realize the hardest part starts after the AI has finished scanning the code.

Legacy estates rarely sit neatly off to the side. They are woven into the operating model of the business. They reflect years of process decisions, integration choices, compliance requirements, customer-specific exceptions and institutional knowledge that is often scattered, tribal or barely documented. Lovely little treasure hunt, except the treasure is risk

An AI model may identify an ageing integration point or highlight an application that supports a critical business process. That is helpful. But the real challenge begins when teams realize how many other systems, workflows and operational teams are connected to what looked like a straightforward change.

In many large organizations, legacy systems are still in place for a very simple reason: they work. They continue to perform reliably under demanding conditions, even if parts of the surrounding environment have evolved, degraded or become harder to support over time.

That is why modernization is not just a technology exercise. It is a sequencing exercise. It is a risk exercise. And, done properly, it is a business decision.

The multi-layer challenge

Every technical decision inside a legacy estate has consequences somewhere else. A change to one application can affect recovery procedures, audit requirements, licensing agreements, batch schedules, integration layers or support processes that have been stable for years.

This is where many modernization programs stall. Teams underestimate how interconnected these environments have become. AI can accelerate the technical assessment, but its real value comes when those insights are connected to the operational and commercial context around the system.

That distinction matters. Enterprises are moving away from broad “replace everything” strategies and becoming more selective. Not every legacy platform needs to be ripped out. Some systems need restructuring. Some need better interfaces. Some need to be moved. And some, frankly, should be left exactly where they are because they are doing their job reliably at scale.

Workload placement has become much more nuanced. Moving a service to public cloud may improve scalability and speed up software delivery, but it can also introduce data sovereignty concerns, latency issues, cost variability or new support dependencies.

At the same time, keeping workloads on modernized IBM Z or Power environments may provide more predictable performance for applications that already run effectively at scale.

The real question is not, “How do we get everything off legacy platforms?” The better questions are, “Which systems genuinely benefit from relocation, which need to be modernized in place, and which can be extended through modern interfaces?”

Without that context, organizations can spend a lot of money moving systems around without actually fixing the underlying problem. Congratulations, you now have the same complexity in a newer location.

We are already seeing this play out in enterprise environments where legacy platforms still sit at the center of high-volume operations. In one recent assessment, AI coding assistants were used to analyze more than six million lines of RPG code running on IBM Power systems, processing roughly 30 million requests a day.

The work surfaced technical debt and concentrated areas of complexity in weeks, giving the organization a clearer basis for deciding what to modernize, where to start and how to sequence change without disrupting core operations.

That is the practical value of AI in modernization: not magic, but better visibility, faster assessment and smarter prioritization.

Why enterprise AI deployments are becoming more specific

This broader shift is also showing up in how hyperscalers talk about enterprise AI adoption. Microsoft CEO Satya Nadella has described the market as moving from “discovery” into “widespread diffusion.” In plain English, the challenge is no longer just building impressive models.

It is embedding AI into real workflows, real operations and real business systems at scale. That is much closer to how modernization actually works inside large enterprises.

The same shift is happening with AI models themselves. The industry still loves to talk about scale, but most enterprise teams are not sitting around hoping for a trillion-parameter model to save them. They need tools that help engineers solve very specific problems inside environments that are already complicated enough.

In many cases, smaller, specialized models are proving more useful because they can be deployed in controlled ways, focused on specific tasks, and governed more tightly.

That governance point matters. Bringing AI into infrastructure operations raises very practical questions: What data can the model access? What systems can it touch? Can it recommend changes? Can it execute them? Who approves movement toward production?

That is another reason task-specific models are gaining traction. Teams can define exactly what the model is allowed to do, where human approval is required and how changes move through existing controls. In enterprise environments, that kind of control is not bureaucracy. It is how you avoid turning a productivity tool into tomorrow morning’s outage bridge.

Where AI is delivering practical value today

The organizations getting real value from AI are usually not the ones making the loudest claims about it. They are applying AI to engineering and infrastructure work that already consumes huge amounts of time: investigating incidents, mapping dependencies, validating changes, supporting regression testing and understanding how complex systems actually behave.

A lot of that work comes down to giving engineers better visibility and helping them get to root cause faster.

AI models can help connect runtime anomalies to recent code changes. They can reduce the time teams spend manually tracing incidents across hybrid environments. They can support regression testing around older applications and surface integration dependencies that were previously difficult to visualize across multiple infrastructure layers.

That becomes especially important in environments where cloud-native services sit alongside long-established mainframe and midrange systems. In many organizations, the hardest problems show up in the seams between those environments, particularly when different teams manage different parts of the estate with different tools, different metrics and different operating rhythms.

That is why the most useful AI deployments tend to focus on practical engineering work, not grand attempts to automate everything at once.

Organizations are seeing value in areas that are repetitive, complex and difficult to scale manually. Automated test generation can reduce regression risk around legacy applications. AI-supported observability correlation can shorten incident investigation cycles. Dependency analysis can help teams prioritize infrastructure work that removes bottlenecks affecting service delivery.

In most cases, AI is not replacing engineering judgment. It is improving the work engineering and infrastructure teams already understand well. And that is where the expectations need to be clear.

AI can absolutely speed up discovery. Work that once took weeks of manual assessment can now happen much faster. But that is usually the point where the real work starts.

A model can tell you how systems connect. It cannot tell you how much disruption the business is prepared to absorb. It cannot decide which customer commitments matter most. It cannot magically unwind 25 years of operational dependency while everyone politely keeps breathing.

Technology leaders should view AI coding assistants as decision-support tools for broader infrastructure and modernization strategies, not as stand-alone solutions to legacy complexity.

IBM’s Bob announcement shows how quickly these capabilities are advancing, especially when it comes to understanding legacy code and helping teams work through large, complex estates. But visibility only matters if organizations can turn it into practical change without creating instability elsewhere.

AI can help you read the legacy estate. It can help you understand the risk. It can help you move faster. But modernization still requires judgment, sequencing and operational discipline.

That part is still very human.

We've featured the best AI website builder.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

A survey found that parents are increasingly going online to find deals in a tough economy.

Here are hints and the answer for today's Wordle for July 13, No. 1,850.

Here are hints and answers for the NYT Strands puzzle for July 13, No. 862.

Here are some hints and the answers for the NYT Connections puzzle for July 13, No. 1,128.

Liene's latest printer is tiny yet mighty and easy enough to use for everyone to enjoy.

• The Nvidia Shield TV (2019) is out of stock pretty much everywhere
• Nvidia says it's down to consumer demand for the streaming device
• It's not yet clear if we'll ever get more stock or a replacement gadget

While there hasn't been any new Nvidia Shield TV hardware for seven years now, the streaming devices continue to be popular — but it looks as though it could finally be the end of the road for the cheapest base model in the series.

As spotted by Android Authority, this $149 / £129 / AU$289.95 model is now out of stock via Nvidia's official channels, and Nvidia has stopped short of promising that those stock levels will be replenished anytime soon.

In a statement to Android Authority, Nvidia said the Shield TV was out of stock "due to demand", and that there was nothing to say "regarding future availability at this time" — so make of that what you will.

Nvidia also took the opportunity to highlight that all Shield TV models continue to get software updates and continue to be supported by Nvidia, even though it's now more than a decade since the first of these gadgets went on sale.

More to come?

The latest Nvidia Shield TV launched in 2019 (Image credit: Future)

Both the base model Shield TV and the more premium Shield TV Pro were given a refresh in 2019, and Nvidia says the Pro version is still on sale. Whether or not Nvidia is planning to let the stock run down on that model too remains to be seen.

I've had several Shield TVs down the years, and have always been impressed with the hardware and the software on offer. They support a wide variety of apps and games, come with local storage, and can be customized in a variety of ways too.

Even something like the recently launched Google TV Streamer only just about catches up to everything that's possible with the Shield TV boxes — and that's saying something considering Nvidia's devices made their debut in 2015.

Nvidia execs are on the record as saying they're open to the idea of new Shield TV hardware in the future, so hopefully this clearing out of stock will be followed by a brand new model in the not too distant future.

The PC gaming landscape has changed dramatically in the last handful of years. We've seen the slow and steady move away from native (purely rasterized) performance and onto the crutch of AI-powered upscaling technology. Whether Nvidia DLSS, Intel XeSS, or FSR 4, Multi-Frame Generation, or "fake frames", have become a core part of the experience.

It doesn't really matter how powerful the best graphics cards are anymore, as AI-powered upscaling has shifted the playable performance expectations across the board. We see this as standard in the system requirements for today's demanding PC games; it's a huge asterisk that developers use to claim otherwise unheard of FPS in intensive software.

Are the likes of Nvidia DLSS, Intel XeSS, and AMD FSR just a failsafe to make up for poor software optimization? That's part of the story, sure, but it's far more nuanced than that. As computing components become more expensive, and AI muscles its way into the territory in a more aggressive manner, the two, which used to go hand in hand, have now become inherently parasitic in a wanton race to the bottom if things are not course-corrected. Here's what AI-powered upscaling means for the future of PC gaming.

Nvidia DLSS 5 is the first symptom of a wider issue

I've been a champion of DLSS for many years, primarily for how it can boost weaker graphics cards to give users playable framerates. It's an ever-evolving AI-powered tech that's continuing to improve and deepen. Some of its best features include Ray Reconstruction, which makes Path Tracing more viable, Frame Generation/MFG, and DLAA for smoother anti-aliasing. When used as an assist to your hardware, it can be the difference between smooth and stuttering, but DLSS 5 is where things simply went too far.

Instead of being a supporting tool, as with DLSS 4.5's Dynamic MFG, the only thing people can take away from DLSS 5 is how AI is actively impacting image quality, and not for good reasons. Described as a "breakthrough in visual fidelity for games", and said to bridge the "cinematic gap", this upcoming AI model uses an algorithm to re-color and overlay motion vectors.

DLSS 5's showcase of results is troubling to say the least. At best, it slightly improves the lighting in EA FC, and at worst, it completely overwrites the distinct visual art style of PC games like Starfield, Resident Evil Requiem, and Hogwarts Legacy. Sure, the lighting is a little better, but it comes at the cost of a flat and artificial-looking brightness of the entire scene, making everything (ironically) look far more lifeless and void of personality.

Nvidia tends to be the frontrunner that AMD and Intel later catch up to. With the DLSS 5 release date still unconfirmed, but claimed to be coming in the autumn of 2026, we (likely) won't see the full ramifications of this on the wider gaming industry until next year, but when this glorified AI-filter drops, it's likely to become an ingrained option in many flagship titles. There's a reason why Team Green started by showing off some of the largest games from the most well-known publishers and developers in the business; if you enforce it at the top, the rest will follow for fear of being left behind.

The frightening reality of how expensive new graphics cards could be

Nvidia's next generation of graphics cards will have AI at the forefront, and they will not come cheap (Image credit: Future)

It's no exaggeration to say that 2026 is one of the worst times on record to build a custom gaming PC. DDR4 and DDR5 RAM prices have skyrocketed due to the global supply of memory modules being drained en masse to build data centers, and the less said about what's happened to flash memory found in the best SSDs, the better. While these individual components doubling in price overnight is already troubling, there are even worse consequences for graphics cards, made all the more infuriating by the fact that it's a self-destructive cycle with (seemingly) no end in sight.

Graphics cards rely on VRAM to have enough bandwidth to perform properly. For Nvidia's current-generation RTX 50 series, that's the superfast, denser GDDR7, whereas AMD and Intel are still using the slower, older GDDR6 standard. As memory modules are becoming scarcer, it massively drives up the price for core components, such as the Samsung, Micron, and SK hynix memory modules needed to build the video cards in the first place.

Put simply, graphics cards will become more expensive because manufacturers are too busy building data centers with the components, meaning the end consumer ends up paying considerably more. We've already seen countless examples of this, such as the Steam Machine's overpriced nature, the Steam Deck's price increase, and even how it's made the PS5 and Xbox Series X more expensive six years in than at launch.

We've seen prices of graphics cards increase massively since the semiconductor shortage, which plagued the RTX 30 series launch. You paid more, you got less, and now the manufacturers know they can overcharge you. In a computing landscape where components can shoot up anywhere from 20 to 50% overnight, it puts the reinforced focus on DLSS, XeSS, and FSR as a necessity rather than an optional helping hand.

The future of AI-powered upscaling is a necessity

The Steam Machine relies on AMD FSR to hit Machine Verified status (1080p at 30 FPS) (Image credit: Valve)

While we're heading towards more expensive graphics cards that rely on AI upscaling tech just to keep up, we can also look at what the future of DLSS, XeSS, and FSR will need to do to keep up. We've already established that DLSS 5 made a contentious call with its AI art filter, but what about real-world, practical uses in 2027 and beyond?

Intel XeSS 3 launched with Multi-Frame Generation, which was rolled out to both its Alchemist and Battlemage graphics cards, even bringing true MFG to handhelds like the MSI Claw 8 EX AI+ and Acer Predator Atlas 8. To look positively at AI-upscaling tech in this regard, its major benefit will be to make handheld gaming PCs, such as a Steam Deck or Lenovo Legion Go S successor, more competitive.

The most recent update to AMD's AI-powered upscaling tech, FSR 4 Redstone, still trails behind DLSS 4.5, which means Team Red will need to strike back with FSR 5 to have a chance at dethroning Nvidia while it is down. AMD's image quality is far better than it used to be, even though its frame pacing leaves a lot to be desired. Based on the track record, from what I've seen from Intel and AMD, XeSS and FSR look to continue to iterate on the core fundamental technologies, whereas Nvidia is looking to do its own thing in counter to what was expected.

The future of AI upscaling tech becomes two-fold. Ideally, enabling any of these settings should be as normal and natural as turning on TAA (temporal anti-aliasing) and forgetting about it. It's easy to forget that there was a time when such a setting was contentious, and now it's almost universally used across the board. It's a similar story to when Nvidia's PhysX SDK was pushed so heavily, as it's now a default setting that's enabled within game engines as standard.

So, AI-powered upscaling is best when it's not noticeable. If you're playing a game and you're noticing smooth performance and a high average FPS, then it's doing its job properly. Problems only really arise when that tech tries to overtake the core experience rather than in support of it, and chiefly as a symptom of a wider problem that we're still experiencing. AI-trained algorithms need data servers; we're making more of those, which means taking away the resources to build graphics cards, meaning you'll pay more for them when the Nvidia RTX 60 series, Intel Celestial, and AMD RDNA 5 eventually roll out.

Will native performance ever be relevant again?

I mentioned above about the asterisk of estimated performance when a game's benchmarks and recommended system requirements go live. Oftentimes, these developer/publisher-approved tables promise 30-60 FPS as standard, and try to discreetly hide that DLSS, FSR, and XeSS are needed to hit that cap. It's something that Valve is just as guilty of with its somewhat questionable claims of 4K60, which was walked back when using FSR.

When the biggest and most well-known entity in PC gaming makes a move like this, the gaming world takes notice, particularly with who Valve was targeting in the first place. If FSR is essential for playable framerates, then it becomes non-negotiable; a forced standard, an excuse for developers to rush out unoptimized games, which have plagued countless PC ports over the last five years.

We find ourselves at a crossroads then. AI-powered upscaling does just as much harm as it does good; it is simultaneously the answer to (and cause of) a fair amount of the problems we're currently experiencing as PC gamers, as the benefits and cons constantly battle out for pole position. It looks as though AMD and Intel are on the right track, even if Nvidia is pacing its own trail, one that (hopefully) isn't followed by its competition. If we're already expected to pay four figures for a "mid-range" GPU, let us hope that it can perform decently enough.

I almost can't write too much of an intro because I'll cry, but Nick (Kit Connor) and Charlie's (Joe Locke) beautiful love story is about to come to a close in Heartstopper Forever.

Since 2022, Heartstopper has captivated the hearts of Netflix subscribers, faithfully adapting the graphic novel series by Alice Oseman of the same name. In fact, if this is all too much for you, you can literally relive the ending again — Oseman has just published its sixth installment.

Frankly, I think Heartstopper will really live up to its "forever" subtitle. But when does Heartstopper Forever arrive on Netflix?

What time can I watch Heartstopper Forever on Netflix?

Heartstopper Forever drops on Netflix on July 17, 2026.

As for exact time, it should be the standard 12am PT release that we saw across seasons 1-3.

For global regions, here's when you need to be prepared:

• US – 12am PT / 3am ET
• Canada – 12am PT / 3am ET
• UK – 8am BST
• India – 12:30pm IST
• Singapore – 3pm SGT
• Australia – 6pm AEST
• New Zealand – 8pm NZDT

When do new episodes of Heartstopper Forever come out?

(Image credit: Netflix)

Heartstopper Forever isn't actually season 4... it's a feature-length film.

According to the streamer, it's clocking in with a runtime of 114 minutes, so that basically the equivalent of 4.5 episodes.

Sadly, this will be it for the show's run on Netflix.

Robotics researchers are trying to prove that lighter-than-air robots could excel at emotional connection.

I doused my Cosmic Orange iPhone 17 Pro in these household chemicals, and the effect was dramatic. Don't try this at home.

Marshall Stanmore IV review

If you want to invite a speaker that looks as good as it sounds into your living room, then the new Marshall Stanmore IV could be a great pick for you. It aims to deliver everything its predecessor did, but better, with a broader soundstage, enhanced bass, and superior controls.

And as someone who owns (and loves) the previous model in this line, best believe I was expecting big things from the Stanmore IV. Luckily, it delivered, offering up impressive sound, a solid set of features, and an eye-catching design. But is it worth its relatively premium price tag? Here’s what I think after many hours of testing.

But before we decide whether the Marshall Stanmore IV can sit alongside the very best Bluetooth speakers, let’s take a look under the hood. This thing essentially uses the same setup as its sibling, the Marshall Acton IV, but crucially with a larger woofer. That means you’re getting a single 5-inch sub with 60W of amplification alongside two 0.75-inch tweeters with 25W of amplification. The result? Big, commanding, and engrossing sound.

I started by firing up Are You Gonna Go My Way by Lenny Kravitz, and the Stanmore IV handled it masterfully. Wailing electric guitars had excellent tonal accuracy and cut through with clarity; vocals also sounded driven and emotive.

Moving over to a deeper track, like Vision of Love by Lewis Taylor, and the Stanmore IV continued to impress. Deep bass pumping through the track hit with tremendous impact while remaining regimented and clean. But thumping low-end never came at the expense of sounds elsewhere in the frequency range. High-pitched drums sounded expressive and vibrant, while vocals in the mid-range were granted plenty of room to play in. Bass can even reach down to 38Hz, meaning you get brilliant, low extension — even sub-bass comes through with vigor and confidence.

LDAC is also on board now for higher-res Bluetooth listening, which helped to illuminate breathy vocal details in Black Eye by Allie X. And even though I’d argue the speaker has a more energetic sound signature, with a lot of focus on the hard-hitting low end, it still supplies a detailed, nuanced listen.

One of the limitations of the Stanmore IV’s little sibling, the Acton IV was its stereo playback. Although it whipped up a decent impression of stereo sound, its small size made it a little difficult to create the most meaningful sense of separation. But the larger, wider build of the Stanmore IV takes things up a notch, and in Jimi Hendrix's Foxey Lady I picked up on a stronger sense of separation, with lead guitar brilliantly placed on the right.

And, more generally speaking, the Stanmore IV whips up a broad, engrossing soundstage. Marshall has improved the tweeters and waveguides on this model to help disperse sound more evenly and provide room-filling sound. However the extra width has been achieved, it certainly gets the nod from me.

Marshall’s Dynamic Loudness tech also ensured that tracks maintained admirable clarity, even at higher volumes. Of course, you can still expect a bit of compression at 100%, say, but I was impressed with the speaker’s control in the upper echelons of loudness. The Stanmore IV is even better in this regard than the smaller Acton IV — perhaps thanks to its larger woofer and larger cabinet size.

Overall, the Stanmore IV provides powerful, driven, and exciting sound, with commendable depth and expression. It's bullish and confident, but still takes time to smell the flowers — and its wider soundstage and refined bass even helps it surpass its already impressive predecessor.

(Image credit: Future)

But now it’s time to move on from sound and look at a few of the features you can enjoy on the Stanmore IV. If you’ve already seen my Marshall Acton IV review, then you’ll get the gist of what’s on board. The Stanmore IV uses the newer Marshall app, enabling you to save three EQ calibrations (using a five-band equalizer), and you can use the M button to cycle between these if you like. The app also opens up placement compensation, enabling you to optimize the speaker’s audio output depending on its positioning, and there’s an option to change the brightness of the LED indicators.

Like the Stanmore III, there’s also RCA and 3.5mm connectivity, allowing you to hook up a turntable, or connect the speaker up to an AUX cable.

Something I wish the Stanmore IV did have is Wi-Fi connectivity. Not only does Wi-Fi provide the highest quality wireless listening experience, but it also prevents pesky sounds from your device — like phone calls and notifications — blasting from the speaker. What’s more, this is a home speaker, so it could easily have a stable connection to your home network at all times.

Marshall has multi-room covered with Auracast tech, which enables a bunch of its speakers to pair together, but I would’ve loved to have seen Wi-Fi onboard for the most seamless, quality-focused listening experience.

Another thing the Stanmore IV leaves out is voice assistant capabilities. Unlike models such as the Sonos Era 100 or Bose Lifestyle Ultra, there’s no smart voice control onboard. Although this is a function that I personally don’t tend to use on speakers, I know that some may wish for it on a model designed primarily for home use.

But something that’s sure to be a hit with most is the Stanmore IV’s design. This thing is an absolute beauty, and although it looks very similar to its predecessor, I’d argue that there’s no need to fix something that’s not broken. The new Stanmore stuns with a gorgeous faux leather exterior, beautiful speaker grille, and luxurious golden detailing. It looks like a true statement piece, and an item that will complement any living space (while still producing excellent audio).

As was the case on the Acton IV, buttons and control knobs are also perfectly responsive and pleasing to use, and there are also onboard EQ controls for altering bass and treble levels if you want to make some changes in a pinch.

So, now we come to the big question. Is the Marshall Stanmore IV worth the money? Well, it’s not the cheapest speaker around, with a price tag of $399.99 / £349.99 / AU$679, making it $100 / £90 / AU$180 more than the Acton IV. On the surface, that may seem like a significant jump for a speaker that’s almost identical — bar a larger cabinet and slightly larger woofer. But these seemingly small changes actually make a significant difference, in my view.

It maintains tighter control at the highest volumes, and also produces the seismic sound that Marshall has become associated with. And that’s not to do the Acton down — it’s just to say that I think you get your money’s worth when stepping up to the Stanmore. I’d also say that the Stanmore competes well against rivals in its price category, with a lower price tag yet more might than a rival like the Denon Home 400. I’d also argue it produces a more striking sound than a model like the Sonos Roam 2 — though you do miss out on Wi-Fi and a few smart features.

Overall, the Marshall Stanmore IV is a great speaker that produces energetic, impactful sound, alongside a stunning look and nifty companion app. Yes, I would’ve loved to have seen Wi-Fi on board, but with LDAC for higher-res Bluetooth streaming added into the mix, I’d still happily recommend this musical maestro from Marshall.

(Image credit: Future)Marshall Stanmore IV review: price & release date

• $399.99 / £349.99 / AU$679
• First released July 2026

The Marshall Stanmore IV was released in July 2026, around four years after its predecessor hit the shelves. This newer model launched alongside the Marshall Acton IV, which is — in essence — a smaller version of the Stanmore. This model comes in at $399.99 / £349.99 (AU$580).

Marshall Stanmore IV review: specs

Weight

8.8lbs / 4kg

Dimensions

13.8 x 8 x 7.3 inches / 350 x 203 x 185mm

Connectivity

Bluetooth 5.3, 3.5mm, RCA

Speaker drivers

1 x 5-inch 60W woofer / 2 x 0.75-inch 25W tweeters

Waterproofing

Not stated

(Image credit: Future)Should I buy the Marshall Stanmore IV?

Attribute

Notes

Score

Features

Multi-room with Auracast, new app works well, LDAC brought in, but lack of Wi-Fi is a shame.

4/5

Performance

Impactful yet detailed audio with tremendous depth, control, and power.

5/5

Design

Very similar to predecessor, but gorgeous amp-inspired aesthetic is massively appealing.

4.5/5

Value

It’s pricey, but stacks up well against competition.

4/5

Buy it if…

You’re focused on getting amazing sound quality
I have to say, the Stanmore IV surprised me by just how good it sounded — even though I already loved its predecessor. Bass is phenomenally powerful yet regimented, mids are driven yet layered, and treble is vibrant yet controlled. Throw in LDAC for higher-res listening and a decently wide soundstage, and you’ve got a great-sounding speaker.

You want a speaker that’s a statement piece
Although the Stanmore IV sounds great, it’s something else that truly helps it to stand out: its design. It maintains that gorgeous amp-inspired aesthetic that’s become synonymous with the Marshall brand, with enticing golden detailing, quality faux-leather casing, and the brand’s iconic logo front a center.

Don’t buy it if…

You’re looking for a portable speaker
When using the Stanmore IV, you’ll need to keep it hooked up to the mains. As a result, it’s better-suited to home use rather than being taken on the road. If you want a more portable option, I’d strongly recommend the Marshall Kilburn III, or awesome non-Marshall alternatives like the JBL Xtreme 5.

You want a smart speaker with Wi-Fi
The Stanmore IV is designed for the home, but it doesn’t have the smart features you’d expect from a rival like Sonos, say. For instance, it leaves out Wi-Fi connectivity (no AirPlay or Spotify connect), which is the most seamless and high-quality way to enjoy music wirelessly. It also leaves out voice assistant compatibility. If those features are important to you, I’d suggest checking out my alternatives below…

Marshall Stanmore IV review: also consider

Marshall Stanmore IV

Sonos Move 2

Klipsch The Three Plus

Price

$399.99 / £349.99 (AU$580)

$449 / £449 / AU$799

$399 / £379 / AU$529

Weight

8.8lbs / 4kg

6.6lbs / 3kg

10.6lbs / 4.8kg

Dimensions

13.8 x 8 x 7.3 inches / 350 x 203 x 185mm

6.3 x 9.5 x 5 inches / 160 x 241 x 127mm

7 x 14 x 8.4 inches / 178 x 355 x 213mm

Connectivity

Bluetooth 5.3, 3.5mm, RCA

Bluetooth 5.0, Wi-Fi, USB-C

Bluetooth 5.3, RCA, USB-C, digital optical

Speaker drivers

1 x 5-inch 60W woofer / 2 x 0.75-inch 25W tweeters

2 x angled tweeters, 1 x mid-woofer

2 x 57mm full-range drivers, 1 x 133mm subwoofer

Sonos Move 2
I’ve used the Sonos Move 2 plenty of times, and I absolutely love it. It plates up gorgeous, detailed audio, alongside seamless Wi-Fi streaming, multi-room capabilities, and convenient voice assistant functionality. It also has a 24-hour battery life, enabling you to take it on the go, and it looks incredibly stylish as well. You can’t ask for much more. Read our full Sonos Move 2 review.

Klipsch The Three Plus
Here’s another stylish speaker that seriously impressed us. Klipsch’s The Three Plus speaker offers assertive and intricate audio, fantastic build quality, and plenty of connectivity options. Read our full Klipsch The Three Plus review.

(Image credit: Future)How I tested the Marshall Stanmore IV

• Spent days testing the speaker
• Used at our dedicated music testing space at Future Labs
• Mainly streamed music over Tidal

I tested the Marshall Stanmore IV over the course of a few days, during which time I listened to hours worth of music and exhausted every feature the speaker had to offer.

Most of the time, I used the Stanmore IV in our dedicated music testing room at Future Labs, where I mainly streamed tunes via Tidal on my Xiaomi 17. To begin with, I sifted through the tracks in our TechRadar reference playlist — which features songs from a wide variety of genres — but I also bumped a bunch of tunes from my personal library.

More generally, I’ve spent years testing audio gear here at TechRadar. I’ve reviewed everything from premium wireless headphones like the Sony WH-1000XM6 through to Dolby Atmos soundbars such as the JBL Bar 1300MK2. I’ve also tested more than 50 Bluetooth speakers, including lots of Marshall models, including the Marshall Middleton II and Acton IV.

• Read TechRadar’s reviews guarantee
• First reviewed: July 2026

Commentary: The Apple Watch is the most accurate wearable I've tested, but its battery life is holding it back.

Generative AI has given rise to a new breed of business that can generate synthetic content, be it video or music, and Suno AI is among the biggest names in this category.

The startup's CEO, Mikey Shulman, however, landed himself in hot water when he made comments about the joy and fulfilment that musicians get from practising their craft.

Making music 'enjoyable' again

Shulman was speaking on the 20VC podcast in January 2025 when he remarked that making music isn't something that most people enjoy doing.

Quote of the day

This article is part of TechRadar Pro's QOTD project to provide an insight into the minds of the brightest and most recognized figures in the technology industry today and in years gone by. Read the full series here.

He framed his words in such a way that would suggest that his platform, Suno, cuts out a lot of these steps – largely centered around mastering the skills that you need – which would, in turn, lower the barrier to entry for those who aren't naturally gifted or have the time to 'get good' at making music.

In his words, he wanted to "[give] everybody the joys of creating music" which he deemed a huge departure from the status quo. His remarks, however, drew the ire of countless working within the music industry as well as regular people on social media.

Struggling artists

The threat of AI to the lives and livelihoods of those working in the music industry is very worrisome, according to research, with workers standing to lose 25% of their income over the next four years. It's no surprise, then, that Shulman's comments instigated such a fierce and violent backlash – forcing the CEO to row back and apologize a couple of months later.

His critics also suggested that the comments fundamentally misunderstood the nature of art and working in a medium such as music, where the hours of toil, practice and refinement are, in and of itself, part of what makes it such a fulfilling endeavor.

However, generative AI is still a new phenomenon and companies like Suno have only just entered the arena. Although the nature of the existential threat to the creative industries is clear, the specific economic impact on artists in the years to come remains unclear – especially in an uncertain landscape in which a strong backlash against AI art is brewing.

Warframe Tau might be Digital Extremes' version of Bungie's Destiny 2 expansion The Final Shape.

• Video showcases China's road-mobile electromagnetic catapult launching an aircraft for the first time
• The footage comes from a now-deleted social media post from the Beijing Institute of Technology's School of Mechanical Engineering
• Leverages China's existing EMALS tech by miniaturizing what it already uses on its aircraft carrier

China seems to have solved one of the biggest challenges modern drones face: deploying without a runway anywhere on the planet, thanks to a miniaturized version of the EMALS catapults found on its newest aircraft carrier.

A short video clip that surfaced towards the tail end of last month, from a social media post by the Beijing Institute of Technology's School of Mechanical Engineering, shows three eight-wheeled trucks linked together in a Lego-like fashion to form a runway that allows a propeller-driven drone to take flight.

The footage also shows all three trucks detached, coupling with each other, and exhibiting all-wheel steering that, in principle, enables them, if given enough space, to launch drones or small aircraft in any direction.

Replacing a runway in the modern battlefield?

China's move is not one that occurred in a vacuum: the United States was the first country in the world to deploy an Electromagnetic Aircraft Launch System (EMALS), sparking an arms race when it launched its first aircraft in 2017 aboard the USS Gerald R. Ford.

China followed suit with its own aircraft carrier, the Fujian, which sports three EMALS catapults to the US's four, possibly accounting for the Chinese carrier's smaller size relative to its principal naval rival's largest.

China has, however, beaten the US to the punch when it comes to demonstrating a portable EMALS, it seems, with its three-truck solution essentially marking a first for either country.

Chinese state-owned giants account for a significant share of the 70+ organizations directly involved in an impressive achievement that sets the stage for the next generation of drone-based combat.

Despite the video making its way across social media, it was taken down by the Beijing Institute of Technology's School of Mechanical Engineering, which originally posted it. The move, however, has not stopped defense forums and Chinese analysts from offering insights into what is clearly a major technological achievement for the country.

With France and India both committed to launching EMALS (albeit on future aircraft carriers for now), only two countries currently have the tech deployed on a ship, and only one of them has a land-based, portable option in play for now: China.

Big news for foldable phone fans this week. Reservations for the next generation of Galaxy Z Fold and Flip series phones have just arrived at the official Samsung Store ahead of the next Galaxy Unpacked keynote event.

The event will take place on July 22, 2026, at 9 a.m. ET / 6 a.m. PT / 2 p.m. BST, which means preorders are likely just a few weeks away. If you're interested in picking up a new device, then you'll find everything we know so far right here, including speculation on pricing and potential preorder deals.

While we're expecting the next generation of the Galaxy Z Fold and Galaxy Z Flip, Samsung has already suggested there's something more in store. In its first teaser for the event last week, the company said that "after seven generations of pushing the boundaries of foldables, Samsung is taking a different approach to how it builds anticipation this year."

That message, coupled with months of rumours about a wider foldable design, has sparked speculation that Samsung could finally be ready to introduce an entirely new type of foldable smartphone. With that said, details are still pretty scant as of writing, so take any speculation here as an educated prediction, rather than the gospel.

You can already reserve a device now

You can sign up to reserve the next Galaxy at Samsung directly at no cost and unlock $30 Samsung Credit towards the preorder of the next device and up to $1,230 off with an eligible device trade-in.

In the UK? Reserve your device right hereView Deal

How much will the devices cost?

Right now, we don't even know what these devices will be called, let alone official pricing, so any speculation here is an educated guess. The Galaxy Z Fold 7 and Flip 7 debuted at $1,999 / $1,299 respectively, so that's a rough ballpark of what to expect. Over in the UK, the last-gen devices debuted at £1,799 / £1,049.

I have a suspicion that Samsung might increase the price for this year's devices, however. RAM shortages and inflation are potential contributing factors here, and I think the brand could potentially increase the price anyway if the upgrades are big enough. If there's a new form factor, for example, I could see Samsung leaning heavily into that as a justification for a price increase.

How much can we expect to save?

(Image credit: Lance Ulanoff / Future)

• Trade-in rebates of up to $1,100 to $1,200
• Additional store credit at Samsung
• Major carriers matching discounts

Samsung itself has confirmed that there will be discounts of up to $1,230 with a trade-in rebate once preorders open. That will most likely cover both unlocked and carrier devices, so that gives you an idea of how premium these devices are likely to be.

Up until now, we've usually seen discounts of up to $1,100 on the latest Galaxy S26 and iPhone 17 series, so it looks like Samsung is being particularly aggressive with its discounts at launch.

Breaking that deal down, it could potentially resemble something like a trade-in rebate of up to $1,100, $100 of bonus credit to spend at Samsung (usually on accessories), and the $30 bonus for reserving a device. That's roughly in line with what the brand has offered over previous launches.

In terms of major carriers, you can definitely expect Verizon, T-Mobile, and AT&T to match those trade-in rebates on their sites. We could potentially see these major carriers waive the trade-in criteria altogether, but they usually only do this when you pair up the device with one of their super high-end plans. We're talking about plans that cost over $60 to $80 a month over a three-year duration, so they're usually pretty damn pricey in their own regard.

What about over in the UK?

You can already reserve a device if you're in the UK too, but annoyingly, Samsung hasn't confirmed any of the opening deals with its reservation page in this region. All we know is that you'll be able to get an extra £100 with your trade-in rebate, plus £30 in store credit if you reserve a device right now.

We've seen trade-in rebates of up to £700 with the Galaxy Z Fold 7 in recent months, so I suspect a rebate of anywhere between £650 and £850 is likely to be the maximum saving on the next-gen device when it arrives. Major networks like EE, Three, and O2 might have higher discounts than this, but those won't be for unlocked devices — you'll probably need a pretty substantial contract to get big savings.

Here are hints and answers for the NYT Strands puzzle for July 12, No. 861.