HomeTagsReinforcement learning

reinforcement learning - Page 19

Stories from reinforcement learning come first, newest first, then stories from connected topics (LLM post-training, AI Models, coding agents, Benchmarks, AI, Open Source), each labelled with where it comes from. This view is not indexed by search engines.

Latest

Shanghai AI Lab Releases Atria Dawn Open Weights With 512K Context

Shanghai AI Lab has posted full Atria Dawn weights on Hugging Face under MIT, with 512K context. Its own table has it trailing Opus 5 on SWE-bench Pro.

Tencent Hunyuan Open-Sources Hy-Nomas, an Interactive World Model

Tencent Hunyuan has released Hy-Nomas under Apache 2.0, a world model that builds video segment by segment from text and camera input. No scores yet.

Ai2 Releases Open Satellite Embeddings for Every 10m Pixel on Earth

Ai2's OlmoEarth v1.3 gives every 10-metre pixel on Earth a 128-number summary, with a free no-sign-in explorer. Its benchmark lead is self-reported.

Anthropic Says Claude Proved 239 Open-Source Entry Points Crash-Free

Anthropic and Theorem used Claude to write machine-checked proofs that 239 OSS-Fuzz entry points cannot crash, finding hundreds of bugs along the way.

Westlake Team Says Its Pinal AI Designs Working Proteins From Text

A Westlake University team says in Nature that its 16-billion-parameter Pinal model turned text prompts into proteins that worked in four lab tests.

Anthropic Gives Agent Skills to AAIF, Saying 70+ Platforms Use It

Anthropic has donated Agent Skills to the Agentic AI Foundation, which also adds OpenAI's Open Responses, OpenEnv and OpenSandbox to reach 10 projects.

Microsoft Diagnostic AI Scored 83% on Past Cases, Science Paper Says

Microsoft's MAI-DxO diagnosed 83% of complex past cases against 36% for doctors in a Science benchmark study. No product, approval or date is announced.

iFlytek Says Its 293B Spark X2.6 Was Trained Only on Chinese Chips

iFlytek says its 293bn-parameter Spark X2.6 model was trained entirely on Chinese chips. It gave no benchmark scores. Only a 4B version has open weights.

Google Cloud Unveils Agent Compute Engine to Run AI Agents in MicroVMs

Google Cloud's Agent Compute Engine gives each AI agent its own hardware-isolated microVM. It reaches preview in the coming months; no GA date was given.

Anthropic’s New Claude API Version Rejects Temperature Settings

Anthropic's Claude API version 2026-10-21 returns 400 errors for temperature, top_p and top_k. New SDKs send it by default and move to Apache 2.0.

Latest on TSN

Explore the TSN Atlas: Every topic we cover, mapped. Click a bubble to dive in.