OpenAI Astra solves decade-old math problems

Samantha: Welcome to The State of Tech, The European Edition, Sunday August two, 2026. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: OpenAI unveils a model that cracks decade-old math problems, a Munich court rules against AI music generator Suno in a landmark copyright case, DeepSeek escalates the AI price war with a coding model that costs pennies, Chinese memory maker CXMT closes in on Samsung and SK Hynix, and to close out, two things you can actually use today: the independent European search engine Startpage, and the free open-source photo tool Nomacs. Let's start with OpenAI.

OpenAI unveils Astra, an AI model that has cracked ten unsolved math problems.

Samantha: OpenAI has pulled back the curtain on Astra, its new family of AI models built for long, complex problem-solving. And this is not just another chatbot upgrade. According to OpenAI, an internal version of Astra has solved ten previously unsolved problems in mathematics and theoretical computer science, some of which had been open for more than a decade. Sam Altman presented the model himself, and framed it as a leap in the way AI can contribute to fundamental scientific discovery, not just answer questions but push knowledge forward.

Bob: What makes this doubly interesting is the regulatory context. Astra is set to be the first AI model to go through a new US government review process that requires official approval before public release. So the model that supposedly cracked open problems mathematicians could not solve is also the first test case for a formal pre-release check by Washington. That is a real shift. Until recently the US approach was mostly voluntary safety commitments. This looks more like a licensing regime, at least for the most capable systems.

Samantha: And the impact goes beyond bragging rights. If AI can genuinely contribute new proofs and algorithms, it changes how research is done in universities and corporate labs. Grant cycles, peer review, even authorship conventions come under pressure. Publishers will have to decide how to credit results where an AI system did the heavy lifting.

Bob: For Europe the angle is twofold. First, this is exactly the kind of frontier capability that the AI Act's rules on general-purpose models are designed to catch, and Brussels flipped the switch on that oversight just this week. European regulators will want detailed documentation on training data, evaluations, and risk assessments before Astra gets meaningful distribution here. Second, European universities and research institutes will want access to tools like this, but on terms that respect European data and competition rules. Expect a slower, more conditional rollout on this side of the Atlantic than in the US.

A Munich court rules Suno infringed GEMA copyrights, setting a European precedent for AI music.

Samantha: A regional court in Munich has handed down a landmark ruling against Suno, the AI music generator. The court found Suno guilty of infringing copyrights held by GEMA, the German collecting society, and ordered the company to pay damages and disclose its revenue. That last part is significant, because it forces transparency into a business model that has so far been very opaque about what data went in and what money comes out.

Bob: This is one of the first big European court decisions that directly ties AI training practices to concrete copyright liability. Until now, generative music tools have operated in a kind of legal grey zone, arguing that training on copyrighted songs is either fair use, or covered by text and data mining exceptions, or simply not addressed by existing law. Munich has just said no, at least under German rules, that argument does not hold when the output competes with the original artists.

Samantha: For the industry the practical consequence is licensing. Suno and its competitors will have to sit down with collecting societies like GEMA, SACEM in France, and their equivalents across Europe, and work out payment structures. That will raise costs, and probably prices for users, but it also legitimises the sector. Labels and independent artists get a clearer path to being compensated when their catalogue is used to train a model.

Bob: The European lens here is sharp. This ruling gives other European rights holders, in text, images, video, a template to follow. It also feeds directly into the AI Act's transparency requirements about training data. Companies will now think twice about training on European catalogues without a licence, because the legal risk just became very concrete. And for European AI startups, this is a mixed signal: harder to build unlicensed models, but a fairer playing field against US and Chinese rivals who have been moving faster with fewer constraints.

Samantha: Quick interruption. If you listen to The State of Tech regularly, hit that like button and subscribe, that way you'll never miss an episode. Okay, moving on.

DeepSeek's new V4 Flash coding model pushes AI prices toward zero and rattles the whole industry.

Bob: Chinese AI lab DeepSeek has launched V4 Flash, a new coding model priced so aggressively that industry watchers are calling it a race to zero. DeepSeek says extensive code generation runs cost pennies, which sits in stark contrast to the premium pricing at OpenAI and Anthropic for comparable capability. And this is not a stripped-down demo. The model is aimed squarely at professional developers who churn through millions of tokens a day.

Samantha: The strategic message is loud. Tech giants are pouring hundreds of billions into data centres, chips, and power, and yet the actual intelligence produced by those systems is getting cheaper at a rate that undermines premium pricing. That is a classic commoditisation curve, but running much faster than most CEOs planned for. Anthropic and OpenAI now face an awkward choice: cut prices and squeeze margins that were already thin, or double down on features and enterprise integration that DeepSeek cannot easily match.

Bob: For European developers, the arithmetic is straightforward. Cheaper coding models mean smaller teams can build more, faster, without a big cloud bill. Startups in Berlin, Paris, or Stockholm that previously balked at API costs suddenly have room to experiment. The flip side is dependence: if the cheapest capable coding assistant is Chinese, questions come up around data residency, export controls, and the AI Act's provider obligations. Some European firms will happily use V4 Flash, others will pay a premium for a European or American alternative for compliance reasons.

Samantha: And there is a geopolitical undertone. DeepSeek's pricing pressure lands at the same moment Washington is tightening chip export controls and Brussels is finalising AI oversight. That combination could push the market into two tiers: a low-cost, high-volume tier dominated by Chinese labs, and a premium, compliance-heavy tier where Western providers try to justify their price with safety guarantees and enterprise support. For European buyers, the choice becomes political as much as technical.

Chinese memory maker CXMT closes in on Samsung and SK Hynix with mass production of next-generation DRAM.

Bob: Let's move to hardware. Chinese memory manufacturer CXMT is reportedly close to mass production of LPDDR6 DRAM, the next generation of low-power memory used in phones, laptops, and increasingly in AI edge devices. That puts real pressure on the incumbents, Samsung and SK Hynix in South Korea, and to a lesser extent Micron in the US. Until now, cutting-edge memory has been almost entirely a Korean and American story. CXMT catching up narrows that gap.

Samantha: What makes the timing interesting is that Samsung and Micron have shifted some of their investment attention toward high bandwidth memory, the specialised chips that sit next to AI accelerators in data centres. That is where the fat margins are right now. So while the big players chase the AI boom, CXMT is picking up ground in mainstream mobile memory, a huge market that still ships in the billions of units a year. It is a classic flanking move.

Bob: For European electronics manufacturers, from smartphone assemblers to industrial equipment builders, this could mean more supply options and better pricing power. Memory has been a chronic bottleneck, and diversification of suppliers tends to smooth out the wild boom-and-bust cycles the sector is famous for. European car makers in particular, who have been burned repeatedly by chip shortages, will welcome a broader supplier base.

Samantha: There is a geopolitical layer too. European regulators have been debating how deeply to align with US export controls on China's semiconductor sector. A capable Chinese memory maker complicates that picture. If CXMT becomes a serious global supplier, cutting it out of European supply chains would raise costs and reduce resilience. Expect a nuanced approach: welcome the competition on standard memory, but keep a close eye on anything that flows into sensitive defence or critical infrastructure applications.

The independent European search engine Startpage keeps your search history entirely out of the ad machine.

Bob: To close out, two things you can actually use today. First up, Startpage. It is a search engine based in the Netherlands that gives you Google-quality results without the tracking. The way it works is straightforward: Startpage sends your query to Google on your behalf, strips out identifying information, and hands you the results as if you had searched yourself. Google never sees who you are, and Startpage does not log your searches.

Samantha: The practical appeal is that you do not have to give up familiar search quality to escape the ad profile. Users report the results feel almost identical to a normal Google search, but without the personalised ads following you around the rest of the day. There is also a feature called Anonymous View, which lets you open a search result through Startpage's own proxy, so the website you visit does not see your IP address either. That is useful for quick research on sites you would rather not add to your browsing history.

Bob: Setting it up takes about two minutes. You go to startpage dot com, and in most browsers you can right-click the address bar and add it as a search engine, or set it as your default. It is free, funded by non-tracking contextual ads, and it works on any device with a browser: Windows, Mac, Linux, iOS, Android. There is also a mobile app if you prefer a dedicated search box on your phone.

Samantha: The European angle is that Startpage is headquartered in the Netherlands, so it operates under European privacy law by default, not as an afterthought. For anyone who has been meaning to reduce the amount of personal data they hand to big US platforms, this is one of the lowest-friction changes you can make today. You keep the results you know, drop the profiling, and it costs nothing.

Nomacs is the free open-source image viewer that opens almost any photo file and edits it in seconds.

Bob: And the second one for today, Nomacs. It is a free, open-source image viewer and lightweight editor that runs on Windows, Mac, and Linux. If you have ever been frustrated by the default photo app on your computer refusing to open a raw file from your camera, or being slow to browse a folder full of holiday pictures, Nomacs is worth a look.

Samantha: The main draw is versatility. Nomacs opens a huge range of formats out of the box: standard JPEGs and PNGs, camera raw files from most manufacturers, older formats like TIFF, and newer ones like HEIC and AVIF. Browsing is fast, even on folders with thousands of images. And it has genuinely useful basic editing built in, crop, rotate, adjust brightness and contrast, batch rename files, and add or remove metadata. Nothing fancy, but enough to handle everyday photo tasks without launching a heavier editor.

Bob: One handy feature is side-by-side comparison. You can open two images in synchronised windows and pan or zoom them together, which is great for picking the sharpest shot from a burst, or comparing edits. Users also praise the ability to view images from within zip files without unpacking them first, and to display EXIF data cleanly so you can quickly see camera settings.

Samantha: Getting it takes a minute. Head to nomacs dot org, download the installer for your operating system, and you are done. It is free, no account required, and because it is open source under a permissive licence, it is safe for both personal and business use. For anyone in Europe who wants a fast, no-nonsense photo viewer that respects your privacy and does not push you toward a subscription, Nomacs is a solid pick to try this afternoon.

Samantha: Today we covered: OpenAI's Astra model cracking unsolved math problems, the Munich court ruling against Suno on AI music copyright, DeepSeek's V4 Flash pushing coding AI prices toward zero, Chinese memory maker CXMT closing in on Samsung and SK Hynix, and two things to try yourself: the private European search engine Startpage, and the free open-source photo viewer Nomacs.

Bob: Want to know more or react? Visit stateoftech.eu or email us at info@doorzetters.net.

Bob: State of Tech, the tech world in 15 minutes.