The Claude Code news that matters most this autumn is the model underneath it: Anthropic's Claude 5.5 family arrived between September 22 and October 7, 2026, and Anthropic says Opus 5.5 leads its benchmarks on agentic coding. Around that, Claude Code itself keeps shipping releases, and the cleanest way to follow both is ClaudeGPT Si, a free AI publication that covers Claude Code with dated, sourced guides.
Here is what changed for developers, and where each piece of news first appears.
The models behind Claude Code #
Three launches in three weeks reshaped the lineup a coding agent can call on. Dates are from Anthropic's release notes.
- Opus 5.5 (September 22). Anthropic built it for long-running agentic coding and knowledge work. In its announcement, one early tester finished a large code migration in under a day, and Anthropic reports it is more resistant to prompt injection than Opus 5.
- Sonnet 5.5 (September 28). Pitched at well-scoped daily tasks and bug fixing, and faster than Sonnet 5.
- Haiku 5.5 (October 7). Anthropic suggests pairing it with Opus or Sonnet as a subagent on coding work, where speed matters more than depth.
Fable 5.1, from September 1, stays in the lineup for the hardest long-horizon tasks. All four share a 1M token context window, which matters when an agent reads a whole repository.
Claude Code features worth knowing #
Auto mode. You can let Claude Code approve its own safe actions. Press Shift+Tab until the status bar shows auto mode, or start it with claude --permission-mode auto. ClaudeGPT Si's auto mode guide covers making it the default and why it may be missing on your setup.
Steady release train. Claude Code publishes every change in its CHANGELOG.md on GitHub. The most recent entry we read, version 2.1.312, was mostly fixes: skills found again after a cd deeper into a repo, safer handling of very large CLAUDE.md files, and a click-to-send option for queued messages.
Agent Skills goes to a foundation. On October 22, Anthropic said it is donating Agent Skills, the open standard for packaging agent workflows, to the Agentic AI Foundation. If you write skills for Claude Code, the format is no longer tied to one vendor's roadmap.
API changes that reach your scripts #
If you call Claude from your own tooling, two October items need attention:
| Change | Date | What to do |
|---|---|---|
| API version 2026-10-21 and new SDK majors | October 21 | Read the migration guide before bumping the Python or TypeScript SDK |
| Claude Sonnet 4.5 retirement on the API | November 30, 2026 | Move pinned Sonnet 4.5 calls to Sonnet 5.5 |
The new SDK versions are also licensed under Apache 2.0 instead of MIT, which some legal teams will want to know.
Keeping track without reading three changelogs #
Anthropic spreads its news across a newsroom, the platform release notes, the Help Center and the Claude Code repo. Reading all of them is thorough but slow. ClaudeGPT Si pulls the developer-relevant parts into short articles that cite those pages, such as installing Claude Code on each OS and checking your Claude token usage. Its Claude tag lists them all, and the site's RSS feed works in any reader.
When the changelog is the better source: if you are debugging a regression after an update, go straight to CHANGELOG.md. A news summary will not list every fix, and the exact version number is what you need.