
Sign up to save your podcasts
Or


OpenAI archives its ChatGPT Retrieval Plugin, ships a PHP vulnerability checker as Codex skills, and Google, Anthropic and DeepMind publish new kits for agents, evals and auditable detectors.
(0:00) - Cold open
(0:36) - openai/chatgpt-retrieval-plugin - source
(2:02) - openai/php2shell-checker - source
(3:01) - google/elephant-goldfish-model - source
(4:01) - anthropics/single-agent-loop-kit - source
(5:03) - openai/MultiLangBench - source
(5:58) - google-deepmind/detectorscope - source
(6:55) - Trend watch
(7:31) - Sign-off
Sources: the labs' public GitHub — every story links its repo above. More at cortech.online/frontier-commits.
Google archives its original Go SDK for Gemini, Google DeepMind opens a Lean repo of Busy Beaver proofs, and Anthropic and OpenAI keep their terminal agents and SDKs on a near-daily release train.
(0:00) - Cold open
(0:17) - google/generative-ai-go - source
(1:34) - google-deepmind/cryptids - source
(2:34) - anthropics/claude-code - source
(3:35) - openai/codex - source
(4:28) - openai/openai-python - source
(5:28) - openai/openai-agents-python - source
(6:25) - Trend watch
(7:06) - Sign-off
Sources: the labs' public GitHub — every story links its repo above. More at cortech.online/frontier-commits.
OpenAI publishes a Lean-checked math archive from an unreleased model, Google's agent mesh goes public, Anthropic opens a model-driven scanner for open-source maintainers, DeepMind ships a protein design pipeline, and Claude Code and Codex keep their release trains running.
(0:00) - Cold open
(0:32) - openai/math - source
(1:46) - google/agentmesh - source
(2:49) - anthropics/oss-scanner - source
(3:55) - google-deepmind/alphaprotein-novo - source
(4:58) - anthropics/claude-code - source
(6:02) - openai/codex - source
(7:00) - Trend watch
(7:32) - Sign-off
Sources: the labs' public GitHub — every story links its repo above. More at cortech.online/frontier-commits.
OpenAI layers ChatGPT plugins on MCP and lets apps sign in with ChatGPT, xAI ships a TypeScript SDK, DeepMind watermarks designed proteins, and Google builds an agent workflow engine with hard stops.
(0:00) - Cold open
(0:40) - openai/mcp-extensions - source
(1:59) - xai-org/xai-sdk-ts - source
(2:56) - openai/sign-in-with-chatgpt-devkit - source
(4:00) - google-deepmind/synthidbio - source
(5:00) - google/lightflow - source
(5:56) - openai/openai-live-console - source
(6:48) - Trend watch
(7:27) - Sign-off
Sources: the labs' public GitHub — every story links its repo above. More at cortech.online/frontier-commits.
Two terminal coding agents ship at a release-a-day pace or faster, DeepMind publishes a p-adic math library, Anthropic's DevRel opens an unofficial Claude Code playground, and new model names surface in OpenAI's SDK.
(0:00) - Cold open
(0:32) - anthropics/claude-code - source
(1:39) - openai/codex - source
(2:38) - google-deepmind/padic-ml - source
(3:33) - anthropics/claude-code-playground - source
(4:36) - openai/openai-python - source
(5:39) - google/adk-python - source
(6:42) - Trend watch
(7:21) - Sign-off
Sources: the labs' public GitHub — every story links its repo above. More at cortech.online/frontier-commits.
Google turns its agent samples into a governed catalog, Anthropic archives its courses and drops a modeling release that was finished on arrival, and two more repos go read-only at OpenAI and DeepMind.
(0:00) - Cold open
(0:40) - google/adk-recipes - source
(2:01) - anthropics/courses - source
(3:01) - anthropics/uplifting-biomolecular-modeling - source
(4:05) - anthropics/claude-for-financial-advisors - source
(5:12) - openai/role-specific-plugins - source
(6:24) - google-deepmind/gemini_icpc2025 - source
(7:28) - Trend watch
(8:17) - Sign-off
Sources: the labs' public GitHub - every story links its repo above. More at cortech.online/frontier-commits.
A proof drop, a raw incident transcript, and an audit tool built so you do not have to trust the lab that wrote it - plus the week both terminal coding agents shipped faster than their own release notes.
(0:00) - Cold open
(0:39) - openai/NavierStokesAndEuler - source
(1:55) - anthropics/mythos-5-incident-transcript - source
(2:56) - openai/openai-deno-build - source
(3:58) - anthropics/claude-code - source
(4:59) - openai/codex - source
(6:00) - anthropics/axt-verify - source
(7:02) - Trend watch
(7:38) - Sign-off
Sources: the labs' public GitHub - every story links its repo above. More at cortech.online/frontier-commits.
Four of this week's six repos exist to constrain, certify, or wrap something else - the frontier labs are shipping plumbing, not models.
(0:00) - Cold open
(0:23) - anthropics/sandbox-runtime - source
(1:46) - anthropics/commerce-agents - source
(2:44) - google/mcp-security - source
(3:51) - anthropics/formal-math - source
(4:50) - google-deepmind/orbit_sfm - source
(5:52) - google-deepmind/agentic_ecology - source
(6:50) - Trend watch
(7:35) - Sign-off
Sources: the labs' public GitHub - every story links its repo above. More at cortech.online/frontier-commits.
No new models this week - just four frontier labs breaking and hardening the interfaces they already shipped, from a coding tool with no source in its own repo to eleven breaking changes in one physics release.
(0:00) - Cold open
(0:42) - anthropics/claude-code - source
(1:58) - openai/codex - source
(2:59) - openai/openai-python - source
(4:00) - openai/openai-agents-python - source
(5:05) - google/adk-python - source
(6:05) - google-deepmind/mujoco - source
(7:13) - Trend watch
(7:52) - Sign-off
Sources: the labs' public GitHub — every story links its repo above. More at cortech.online/frontier-commits.
From the publisher's feed