Model T

A new open-source project called Strata claims to run Qwen 3.8 Flash Next, a 125-billion-parameter model, on a single consumer RTX 4090 graphics card at roughly 100 tokens per second. The claim has drawn strong interest from developers, who are discussing performance figures, memory requirements and whether the results hold up in practice.

A developer has released an open-source model routing tool designed to give coding agents performance on par with Astra-level systems. The project routes requests between models to optimise quality and cost, and is being shared for public use and feedback.

A developer has published a write-up after spending one month coding with GLM 5.3 Flash, an AI model used as a programming assistant. The post shares a month of hands-on experience with the model for software development work, and the write-up is drawing attention and discussion among developers weighing how newer AI coding models perform in day-to-day use.

OpenAI and Synopsys have announced GPT-Synopsys, described as frontier intelligence aimed at revolutionizing chip design. The partnership pairs OpenAI's models with Synopsys's electronic design automation tools, suggesting AI could accelerate how semiconductors are engineered. The announcement is drawing attention among engineers and technologists debating how far AI can go in automating complex silicon development.

Searches for ChatGPT login are trending in the US and India, with over 1,000 engagements. The spike comes as OpenAI rolls out new features that directly challenge the traditional app store model, prompting renewed user interest in accessing the chatbot. TechCrunch reports the latest moves position ChatGPT as a platform in its own right.

Strata, a new tool presented on Hacker News, describes itself as an expressive semantic layer that can refuse to answer queries from large language models. The pitch is aimed at developers building AI features on top of databases, where control over what an LLM is allowed to ask matters. Commenters are weighing in on the launch and the broader question of constraining AI access to data.

A claim circulating online says climate models have missed something significant, warning that billions of people could be affected. The headline gives no specifics about what was missed, which models are involved, or what the consequences would be. No scientific study, institution or named researcher is cited in connection with the claim, and no independent reporting on it is included in the available material.

A Red Hat developer article benchmarks AI-based decision models, including one called Jev, against LLM-as-a-judge setups and traditional classifiers used as guardrails. The reported finding is that the decision models do not outperform either alternative, suggesting simpler established approaches remain competitive for automated decision and moderation tasks.

A 3D artist has published a detailed breakdown of creating a lived-in stylized Viking environment, walking through the modeling, texturing and scene-dressing techniques used to give the settlement an authentic, weathered feel. The breakdown covers the full workflow from blockout to final render, and is drawing attention from game artists and environment designers looking for tips on stylized world-building.

A Wall Street Journal report argues that companies are struggling to plan budgets for AI, because usage-based pricing such as per-token costs, rapidly changing model prices and unpredictable adoption make spending volatile and hard to forecast. The piece has struck a chord with business and technology readers debating how firms should manage costs that scale in unpredictable ways.

An Atlantic piece argues that AI's biggest contribution to science is not replacing scientists but sifting through vast mountains of biological data, allowing researchers to work faster. Advanced models are described as accelerators for biological research, handling scale and pattern-finding at a level impossible for humans, while scientists retain judgment and interpretation of results.

A developer has released Minigraf, an open-source, embedded graph database written in Rust with bi-temporal capabilities, meaning it tracks both when events occurred and when they were recorded in the system. The project is available on GitHub, and early discussion on Hacker News is focused on what a bi-temporal data model offers for auditing and time-travel queries in embedded use cases.

A rigorous re-examination of cosmic gamma-ray observations has found no evidence that Lorentz invariance breaks down at high energies. Researchers applying unified statistical methods confirmed that the speed of light remains constant regardless of photon energy or direction, reinforcing Einstein's relativity and placing tighter limits on quantum-gravity models that predict energy-dependent light speeds.

Physicists have outlined how a photon with an energy of 300 teraelectronvolts — among the most energetic gamma rays ever detected — could have travelled to Earth. The finding addresses a puzzle in high-energy astrophysics, since such photons should normally be absorbed along the way, and it points to extreme cosmic accelerators capable of producing them.

Researchers at the Wellcome Sanger Institute are using mathematical models to study how cells communicate with each other. The approach aims to unpick the complex signalling between cells, helping scientists better understand tissue function and disease processes.

Researchers have introduced a new AI model that reads protein clues in three different ways to map how proteins contact each other inside cells. The approach could help scientists better understand cell biology and disease mechanisms. Details about the team, data and validation remain limited in early coverage, but the claim of a multi-angle protein analysis method is drawing attention in the biology community.

The Guardian has published its review of Apple's iPhone 18 Pro, judging that the handset brings substantial internal changes while offering only modest practical differences for everyday users. The verdict suggests the new model is technically significant but may not feel transformative to buyers upgrading from recent versions.

A new article examines the black hole information paradox, explaining why quantum unitarity clashes with Hawking's thermal radiation and how the Page curve frames the problem. It also covers recent island and replica-wormhole calculations that reproduce the curve in simplified models, along with the firewall argument and the ER=EPR conjecture linking entanglement to spacetime geometry.

Anthropic has published a new piece of research titled 'Claude-Shaped Science', examining how its AI assistant Claude is being used in scientific work and how it may be influencing the way research is conducted. The publication is drawing attention among technologists and researchers, with discussion focused on what it means for the future of AI-assisted science.

Researchers have unveiled a model combining physics-based modeling with artificial intelligence that reads muscle fatigue signals to support rehabilitation for lower back pain. The approach aims to give clinicians a more precise way to monitor patient recovery. Details on the research team and validation results remain limited.

Researchers report that systems previously treated as fundamentally unpredictable at the quantum level show recurrent patterns when observed over time. The finding suggests hidden structure within quantum chaos, which could sharpen how physicists model complex quantum systems. Details of the study, the team behind it, and the practical implications have not yet been widely reported, leaving much of the reaction so far limited to the announcement itself.

Researchers have proposed a new algorithm capable of reconstructing fluid flows from limited data, according to a recent report. The method could allow scientists and engineers to model complex flow behaviour without dense measurements, which is valuable in fields like aerodynamics, weather prediction, and industrial design. Discussion of the technique is still at an early stage.

Astronomers have identified J0313-1806 as the most distant quasar yet discovered, observed just 670 million years after the Big Bang. The quasar is powered by a supermassive black hole exceeding 1.6 billion solar masses, an age and size that challenges current cosmological models explaining how such massive objects formed so early in the universe.