#Reasoning

AINoam Brown reposted an OpenAI statement that an internal model group reached a Navier–Stokes solution in 88 hours using about 10,000 coordinating AI agents. OpenAI said the model shows a step-function improvement on many benchmarks and that its training is ongoing, with monitoring and isolation safeguards applied throughout. The attached chart compares GPT-6 Astra and the internal model on a curated set of open math problems across test-time compute levels, with the internal model scoring higher at each point.

Why it matters: The quoted OpenAI post gives concrete figures on an internal model's Navier–Stokes result and on a benchmark comparison, showing how the model performs on open problems.

AITencent Hunyuan says its Hy4 preview has been upgraded to reduce long thinking and over-verification on complex tasks, which users had flagged. The company reports the same task quality with fewer turns and lower input and output tokens, confirmed by benchmark and human evaluation. The upgrade is live for all users, and Tencent says it will keep iterating based on feedback.

AIOpenBMB released JustRL-II-base-model, the pre-RL starting checkpoint for the JustRL II math-reasoning case study, scoring about 61% on AIME 2025 before reinforcement learning. The full JustRL II recipe reaches 81% on AIME 2025 in about 300 RL steps from this checkpoint, versus about 74% for a standard GRPO baseline. The Llama-architecture weights are available on Hugging Face and are intended for reproducing the recipe and research on long-CoT RL, not general assistant use.

AIJohn Schulman says a metric for explanation quality, centered on counterfactual simulatability, enables hillclimbing, and praises Adam et al. for a more diverse and realistic dataset and pipeline. He notes that models can be trained to write better post-hoc explanations of their own behavior, as described in a linked thread by @a_karvonen. That thread reports training on thousands of self-explanations of in-the-wild behaviors, with generalization to held-out evals.

AISebastien Bubeck highlights Astra's TikZ unicorn drawing, which he calls remarkable for a single entity. He says Astra scored essentially 100% on FrontierMath Tier 4, ARC-AGI-3, and ExploitBench, and that anyone can access it publicly.

AINoam Brown, speaking for OpenAI, says he is most excited about GPT-6 Astra's potential for scientific discovery and says OpenAI has not yet pushed the model to its limits on math and science. He looks forward to seeing new scientific breakthroughs built with the model. Background context from a quoted post notes a new OpenAI repo containing a Lean formalization by GPT-6-Astra that proves infinitely many pairs of consecutive primes are at most 186 apart.

AIOpenAI's GPT-6 Astra (max) scores 62.7% on ARC-AGI-3 Semi-Private for $26K under the Standard harness, and 99.9% for $19K under the Provider Adapter harness. The authors say Astra used fewer actions than the human baseline on 96.0% of levels, and they note it is not claimed to be AGI.

Why it matters: The report pairs benchmark scores with replays of the model's notation and tool use, showing how it solved unfamiliar environments rather than only that it did.

AINVIDIA has published Nemotron-3-Labs-Ultra-Math-RL on Hugging Face, a 550B total, 55B active parameter model for solving difficult math problems and identifying proof mistakes. The model is part of an ensemble that reached gold-medal level at the International Mathematical Olympiad 2026, and it is available for commercial and non-commercial use under the OpenMDW-1.1 license. Deployment is designed for NVIDIA Blackwell or Hopper GPUs, with a recommended minimum of 8× B200 on a single node and a context length of up to 1M tokens.

Why it matters: The release details the model's math-proof role, its 550B total and 55B active parameters, and its vLLM deployment requirements for teams weighing adoption.

AIGoogle introduces Gemini 3.8 Flash for coding, agentic tasks, and multi-step reasoning, priced at $0.75 per million input tokens and $3.75 per million output tokens during the introductory period. Gemini 3.8 Flash Cyber targets vulnerability detection and automated patching and is available only to trusted defenders through the new Fairwind Program. The introductory price expires December 31, 2026, after which $1.50 and $7.50 per million tokens apply.

Why it matters: The post separates a general coding and agent model from a restricted cyber variant, showing how one shared core is deployed under different access and safety tiers.

AIGoogle's RT-2 model, announced in July 2023, trained a multimodal LLM to output robot actions directly, and the article argues this approach launched the current robotics boom. The author follows later work from Physical Intelligence, including action chunking with flow matching, reinforcement learning on real robots, and visual subgoal generation, and notes that the field is debating whether vision-language-action models will give way to world models.

AIwith gains over 3.7 Flash in software engineering, agentic tasks, and multi-step reasoning. Sundar Pichai says it outperforms most larger frontier models on DeepSWE v1.1 at a fraction of the cost. The comparison table lists input at $0.75 and output at $3.75 per 1M tokens, with introductory pricing through December 31, 2026.

AIGoogle AI Developers says Gemini 3.8 Flash, built for complex reasoning, generated an interactive 3D visualizer using Three.js in Google AI Studio. The visualizer produces physically proportioned teardowns of hardware devices, automatically splitting each device into layers that users can explode and inspect with a deconstruction slider.

AIGoogle's Koray Kavukcuoglu says Gemini 3.8 Flash is a major step up from Gemini 3.7 Flash and outperforms most larger frontier models on complex engineering problems at a fraction of the cost. The attached DeepSWE V1.1 chart, sourced to Datacurve AI, plots average cost per task against score for Gemini 3.8 Flash and other models. A link to Google's blog post with more details is included.

AIGoogle AI Studio announced Gemini 3.8 Flash, which it calls its most intelligent workhorse model. The company says it brings significant improvements over 3.7 Flash in software engineering, agentic tasks, and multi-step reasoning in specialized domains. It is available at the same introductory price as 3.7 Flash, $0.75 per million input tokens and $3.75 per million output tokens, through the Gemini API and AI Studio.

AICohere Labs released Tiny Aya En-Thinker, an open-weights 3.35 billion parameter multilingual reasoning model with a 32K context length. It is trained on English reasoning traces for 44 languages plus English, with coverage extending to 20+ more languages through non-reasoning instruction data. The model is available under a CC-BY-NC license that also requires adherence to Cohere Labs' Acceptable Use Policy.

AICohere Labs released Tiny Aya L2-Thinker, an open-weights 3.35 billion parameter multilingual reasoning model that thinks in the same language as the user's prompt before answering. The model supports in-language reasoning for 44 languages plus English, with coverage extended to 20+ more languages through additional non-reasoning instruction data, and has a 32K context length. It is licensed under CC-BY-NC and is available on Hugging Face.

AISebastian Raschka argues that the looped transformer approach attributed to OpenAI's Astra is a minor architectural tweak, not a major breakthrough. He explains that Nanbeige4.2-3B reuses its 22-layer stack twice, effectively doubling depth without adding weights but roughly doubling compute, and that the idea traces back to the Mixture-of-recursions NeurIPS paper. He adds that layer reuse does not inherently hide chain-of-thought, though it could shift more computation into latent activations.

AIOpenAI's Jakub Pachocki says the computation graph depth of current frontier models, including Astra, is within a factor of two of GPT-4. He adds that chain-of-thought monitoring, which OpenAI has used since its first reasoning models, is fragile and trending negatively, though the company is researching ways to strengthen it.

AIAnthropic has released Claude Fable 5.1, the latest upgrade to its most capable class of models, and it is available everywhere today. The company says it handles complex, long-running, multi-step work and avoids shortcuts when fixing root causes of software issues. At lower effort levels, Fable 5.1 can match or beat Fable 5 at a much lower cost, according to Anthropic's benchmarks.

Why it matters: The source names the upgraded model class and its cost tradeoff at lower effort levels, which helps readers weigh it against the earlier version for their own workloads.

AIAnthropic has released Claude Fable 5.1, an upgrade to its most capable model class, and says it is available everywhere today. The company reports that at lower effort levels, Fable 5.1 can match or beat Fable 5 at a much lower cost. It is described as strong at complex multi-step work, such as long proofs and contracts with hundreds of cross-references, and at fixing root causes in software issues.

Why it matters: The source reports cost and effort-level tradeoffs for long-running tasks, helping readers judge whether the upgrade changes their workloads or budgets.

AIAnthropic has launched Claude Fable 5.1 and Claude Mythos 5.1, which it describes as the world's most advanced models for coding and knowledge work. The release notes link to a blog post with more details, but the notes themselves give no benchmarks or specifications.

Why it matters: The source names two new model versions and points to a companion blog post, so readers can compare the release details there.

AIFireworks AI announced general availability of its Training API, which connects a customer's Python training loop to managed distributed training and rollout infrastructure. Serverless training bills per token for LoRA adapters, while Dedicated training provides per-GPU-hour capacity for full-parameter runs and larger models. The post cites customer results, including Heidi moving a clinical scribe from proof of concept to production in four weeks with 3.5x lower latency.

AIAlibaba-NLP has published Core-Reranker-8B on Hugging Face, an 8B-parameter multimodal reranker fine-tuned from Qwen3-VL-Reranker to better distinguish attribute-object bindings in text and image relevance scoring. On compositional reasoning benchmarks COLA, SugarCrepe++, and NegBench, it reports an 82.7% total average, 10.7 points above Jina-Reranker. The model is part of the Core-Embed family, which also includes 2B and 8B embedding models, with Core-Embed-8B reporting a 0.666 total average.

AISebastian Raschka shares a short video explaining the relationship between conventional LLMs, reasoning models, and agents. It also discusses "from scratch" approaches to building them and shows how to install Python and PyTorch requirements with uv.

AIAmazon Science proposes dependence-aware aggregation using Ising models to account for shared prompts, model families, or training lineage among LLM judges. The post says this approach improves accuracy by 9–14% over weighted majority vote.

AISoumith Chintala argues that once you know the tasks you care about, customizing a model beats using a general one. The post is brief and offers no benchmark figures, but it is supported by the referenced Tinker work, where RLVR with expert judgment produced a text-to-SQL model that beat the human baseline.

AIPollen Robotics has unveiled Microduck, a 25 cm open-source biped with 15 actuators and sensors including a camera, speaker, and LiDAR that users can train with reinforcement learning. The robot ships with more than half a dozen pre-trained policies for walking, sitting, roller-skating, and picking up objects with its articulated beak, and costs less than $400.

AIGoogle DeepMind VP of research Zoubin Ghahramani, a Cambridge professor, discusses how machines can represent uncertainty, a line of work he has pursued for about 30 years. The video covers correctness versus confidence, Bayesian thinking in AI, and whether improving machine uncertainty is a missing piece for future AI progress.

AIAn anonymously posted elliptic curve has been identified as having rank 30, a notable result in number theory. OpenAI's Kevin Weil highlights it as a case of humans and AI learning new mathematical facts together.

AIZhang Yijia, founder and CEO of Chinese tech think tank 甲子光年, presented a 2026 AI trend report at a Beijing investment conference, arguing China's venture capital has entered a new cycle as financing, investment, and exits rebounded in H1 2026. He said AI absorbed over 70% of global venture investment in H1 2026, with OpenAI and Anthropic together raising $217 billion, and that global AI-related capex is expected to exceed $1 trillion in 2026.

AIJason Wei now believes a small 1B-parameter "cognitive core" relying on tools is insufficient, because fast, natural recall without tool use matters. He cites speed, knowledge better learned through backpropagation than retrieved from search, and the greater reliability of already-known facts over repeated lookups. Since a 1B model has an information limit, he argues that demanding AI will still need larger models, not just tool access.

AIDiG-bench, a 70-game benchmark for discovering hidden rules through interaction, shows Opus 5 and Fable 5 with Claude Code performing best overall, with GPT-5.5 next. Only Opus 5 and Fable 5 beat any Tier 7 tasks, at a 0.2 success rate, while humans reached 100% on the same tests. The authors say the benchmark's games are mostly kept private to avoid training contamination.

AII find the "inception" pattern to be very useful in many agentic use cases. You can force the model to take an action when it thinks for too long by injecting a thought after a specified reasoning budget. Helps dealing with underspecified tasks which make the model reason for way too long.

AIEpoch AI outlines nine open questions about AI capabilities, including whether AI can take over full jobs and whether benchmark scores are correlated. The author says Epoch's benchmarking work is built to help answer them, citing examples such as MirrorCode, Remote Labor Index, and the Epoch Capabilities Index (ECI). The post notes that benchmark scores are highly correlated across domains, and that ECI growth trends can help detect whether AI capability progress has accelerated.

AIDemis Hassabis says Gemini 3.7 Flash is also lightning fast. Logan Kilpatrick's post identifies the model as 3.7 Flash, with no speed figures given.

AIGoogle has introduced Gemini 3.7 Flash, linking to an official blog post for more details. The post itself provides no further specifications, benchmarks, pricing, or availability information.

AIDeepSeek announced the launch of DeepSeek-V4-Pro, citing major Agent upgrades and flexible reasoning effort settings of low, high, and max for V4-Pro and V4-Flash. The model supports the native OpenAI Responses API and is optimized for Codex with one-click setup. V4-Pro is available on the app and web through Expert Mode and via API, with model names unchanged.

AIA neurosurgery resident at Peking Union Medical College Hospital, Shanmu Jin, posted a preprint claiming a proof of Crouzeix's conjecture in numerical linear algebra after using ChatGPT 5.6. The essay by Alex Townsend and Anne Greenbaum says the conjecture had been open for more than two decades and that the argument held up after a few hours of review. Bubeck, who says he spent a week on the problem in 2012, quotes the story and calls it amazing.

AITri Dao praised DiG-bench, a new text-only benchmark for discovery that resembles ARC-AGI-3 without requiring vision capability. The benchmark, built by researchers from Princeton, MIT, KAUST, and Inria, tests frontier models on text-based discovery games. Their early findings indicate frontier models have improved substantially but still struggle with some surprisingly simple problems.