Releases Index

Frontier AI Labs

Frontier AI research labs and model providers, side by side.

This week's digestOct 26 – Nov 1, 2026

GPT-6 goes to the life sciences as the SDKs converge on agents

OpenAI put GPT-6 behind biology-specific safeguards with GPT-6 Rosalind Research, rolling it out across Work, Codex, and the API for approved life sciences teams, while Codex and the OpenAI SDKs spent the week building out agent orchestration, and Claude Code hardened its permission model.

Read the digest45 releases covered
In this issue
01GPT-6 goes to the benchModelsModelsChatGPT Enterprise & EduChatGPT Enterprise & EduChatGPT BusinessChatGPT BusinessAnthropicAnthropic4 releases02Agents become the unit of work in the SDKsOpenAIOpenAI17 releases03Codex and Claude Code harden the local agent loopCodexCodexClaude CodeClaude Code6 releases04Product surface and platform plumbingChatGPTChatGPTModelsModelsChatGPT Enterprise & EduChatGPT Enterprise & EduAPIAPIChatGPT BusinessChatGPT BusinessClaude PlatformClaude Platform12 releases
EarlierOct 19OpenAI and Anthropic both bet on cheaper, longer-lived contextOct 12GPT-5.5 retires, Claude Code retunes permissions, and the SDKs rebuild for realtimeAll digests →
Week of Nov 2in progress · digest Monday
FriNov 6, 20263 releases
AnthropicOpenAI
Daily summary

Claude Code closes a sandbox escape and hardens auto mode

Claude Code tightened auto-mode handling so calls the server can't check are asked about or denied rather than silently falling back to the built-in classifier, and patched a worktree sandbox escape plus a Bash permission bypass that could hide part of a command.

  • Claude Code no longer falls back to its built-in classifier when telemetry or feature flags are off, closing a gap where unchecked calls could run; CLAUDE_CODE_AUTO_MODE_SERVER=0 restores the old behavior.
  • Claude Code fixed a sandbox escape during worktree creation and a Bash permission bypass where a crafted environment prefix could conceal part of a command from checks.
  • Claude Code added terminal support for MCP tasks, letting servers run long operations through the session.
  • OpenAI's Node SDK now surfaces agent session errors as stream events instead of only through the final session outcome.
AnthropicAnthropic2
Claude Code

Auto mode no longer falls back to built-in classifier; sandbox escape fixed

v2.1.319

Auto mode with telemetry or feature flags off now asks about or denies calls the server can't check instead of falling back to the built-in classifier, with CLAUDE_CODE_AUTO_MODE_SERVER=0 restoring the old behavior. The release also fixes a sandbox escape when creating a worktree and a Bash permission bypass where a crafted environment prefix could hide part of a command from permission checks.

Details
OpenAIOpenAI1
Node SDKv7.53.0
ThuNov 5, 20266 releases
OpenAI
Daily summary

OpenAI launches GPT-6 Cyber for finding and fixing vulnerabilities

OpenAI released GPT-6 Cyber, a cybersecurity model trained to discover exploitable vulnerabilities and develop fixes within authorized scope, while Codex CLI added an agent command center with unread-result markers and cross-backend session switching.

  • GPT-6 Cyber is a dedicated security model for locating exploitable vulnerabilities and producing fixes, trained to stay within authorized scope
  • Codex CLI opens an agent command center via codex agents --no-daemon, marking unread results and letting the session picker switch between backends without losing drafts
  • The Responses API across the Go, Node, and Python SDKs now accepts a model option for specifying the model per request
OpenAIOpenAI6
Codex

Agent command center lands; sessions switch across backends

v0.167.0

Codex CLI gains an agent command center via codex agents --no-daemon that marks unread results, and the terminal session picker can now switch between connected backends while keeping drafts. Saved chats stay visible after switching model providers, network approval prompts close when their requests end, and Ctrl+L clears the display while keeping history available through Ctrl+T.

Details
Go SDKv3.96.0
Node SDKv7.52.0
Python SDKv3.47.0
.NET SDKOpenAI_2.15.0
WedNov 4, 20266 releases
OpenAIAnthropic
Daily summary

OpenAI lists GPT-6 Rosalind across SDKs and widens data residency

OpenAI added GPT-6 Rosalind to the model catalog in its Node, Go, and Python SDKs alongside new data residency options for Australia, India, and Japan, while Daybreak Red brought GPT-5.6 Cyber to AWS GovCloud US-East through Bedrock for approved cybersecurity customers.

  • GPT-6 Rosalind is now callable through the OpenAI Node, Go, and Python SDKs, which also gain Australia, India, and Japan data residency options.
  • Image requests across the SDKs gained complexity controls plus GIF and SVG output options.
  • Daybreak Red made GPT-5.6 Cyber available to approved customers in AWS GovCloud US-East via Amazon Bedrock for authorized cybersecurity work.
OpenAIOpenAI4
Node SDKv7.51.0
Go SDKv3.95.0
Python SDKv3.46.0
AnthropicAnthropic2
Claude Platform

Org IP allowlist; admins can sign members out everywhere

Organizations can now configure an IP allowlist in the Claude Console of up to 1,000 IPv4/IPv6 addresses or CIDR ranges, refusing requests from unlisted addresses. The Admin API can sign a Claude Enterprise member out of Claude everywhere while keeping them in the organization.

Details
TueNov 3, 20265 releases
OpenAIAnthropic
Daily summary

OpenAI SDKs add 12-hour Responses caching and workload identity

OpenAI's Node and Python SDKs extended Responses caching to a 12-hour retention window with cache-write usage reporting, while all three SDKs gained workload identity providers and federation rules and project service tier settings; Claude Code added sign-in expiry visibility and fixed data-loss bugs.

  • OpenAI Node and Python SDKs extend Responses caching to 12-hour retention and now report cache-write usage, giving callers visibility into what cached traffic costs.
  • Go, Node, and Python SDKs all added workload identity providers and federation rules, opening the API surface for credential federation across workloads.
  • Project service tier settings are now configurable through each SDK rather than only at the account level.
  • Claude Code surfaces stored claude.ai sign-in expiry in auth status and JSON output, and fixes config/login loss on shared network home directories plus a --continue regression that dropped earlier conversation history.
OpenAIOpenAI4
Go SDKv3.94.0
Node SDKv7.50.0
Python SDKv3.45.0
AnthropicAnthropic1
Claude Code

Sign-in expiry surfaced; shared-home config no longer silently dropped

v2.1.318

Stored claude.ai sign-in end dates now appear in claude auth status and claude -p JSON results, with a stderr warning starting 10 days before expiry. Fixed config and login loss on shared network home directories, and a regression in 2.1.314 where --continue could lose the earlier conversation.

Details