Track model launches, papers and technical shifts across fields. Sort the same feed by impact, public attention or practical value — and always inspect the source.
Atlas, robotics simulation and a clearer renderer–simulator–planner taxonomy are turning “world model” from a loose label into an actionable stack.
The frontier is moving from impressive robot demos toward scalable training loops: simulation, real-to-sim-to-real data, world action models and reliability evaluation.
Attention is shifting from general agents to the missing execution layer: verified skills, domain tools, reproducible environments and evidence-aware workflows.
Fast retirements make model availability a research-method issue. Reproducibility now requires snapshots, migration evals and explicit product-versus-API status.
No official Codex release is in the current snapshot yet.
No public Codex discussion is in the current snapshot.
Community engagement stays platform-local and is neither a benchmark nor a product-performance claim. Only direct OpenAI or openai/codex links receive the verified label.
Counts are items tracked by this feed, not the total number of papers on either platform. HF votes are a community signal, kept separate from scientific impact.
Countdowns use the official deadline date. Open the organizer page for the authoritative time zone and policy.
Astra is rolling out across ChatGPT, Codex and API surfaces. OpenAI reports large gains in computer use, coding, science and cybersecurity; benchmark figures remain vendor-reported until independently reproduced.
Anthropic recommends moving workloads to Claude Opus 4.8. It is a timely reminder to pin model migrations to evals, not model names alone.
GPT-5.1 Instant, Thinking and Pro left ChatGPT after GPT-4o, GPT-4.1 and o4-mini. API availability follows a separate lifecycle.