OpenAI is developing 'Astra', a multi-agent model family designed for long-horizon, collaborative problem solving over hours or days.
DeepSeek released V4-Flash-0731, a 304B parameter model boasting significantly upgraded agentic capabilities.
Supabase open-sourced an Apache-2.0 benchmark to evaluate coding agents on real-world database, schema, and RLS policy tasks.
ByteDance's Seedance 2.5 generates unified 30-second video and audio clips, tripling the duration of competing models.
Google introduced Managed Agents for Gemini 3.6 Flash, adding native hooks and triggers for agentic workflows.
Anthropic published its official policy stance on open-weights models, balancing safety concerns with innovation benefits.
OpenAI published solutions to ten long-standing open problems in mathematics and theoretical computer science.
A new lightweight evaluation suite, smevals, has been released for testing models, prompts, and agent harnesses.
Datasette-agent 0.4a0 introduces a mechanism allowing agent tools to execute code directly in the user's browser.
OpenAI has reduced the context window size of its Codex model from 372k down to 272k tokens.
Google expanded its robotics lineup by introducing two new physical platforms, Duo and Apollo, powered by Gemini ER 2.
A new field report documents how researchers are deploying AI coding agents to modernize legacy scientific computing codebases.