LLM-Anthropic Plugin 0.27 Updates for Python Library 1.0

👁 118 reads

Simon Willison released llm-anthropic 0.27 on August 24, 2026, updating the Anthropic plugin for his LLM command-line tool to maintain compatibility with the newly released anthropic v1.0.0 Python library - a library update that swaps the longstanding httpx HTTP client for its successor, httpx2.

What changed in anthropic v1.0.0

The anthropic v1.0.0 Python library is a breaking release. Its headline change is the replacement of httpx with httpx2 as the underlying HTTP transport layer. For developers who depend on the Anthropic SDK, this means any code or plugin that previously worked against the pre-1.0 library needs updating before it can function correctly against the new one.

Anthropic published a dedicated migration guide covering the steps required to upgrade to v1.0. The guide is hosted in the official anthropic-sdk-python repository on GitHub and walks through the dependency and API surface changes involved in the transition.

The same shift happened at OpenAI roughly two weeks earlier, when the OpenAI Python library made the identical move to httpx2 in its own v3.0.0 release. The parallel timing suggests that httpx2 has reached a level of maturity that the major AI SDK maintainers are converging on, and that plugin authors across the LLM ecosystem are now working through a similar wave of compatibility updates.

How Willison approached the migration

Rather than making the changes manually, Willison used the task as a practical test of AI-assisted development. He describes prompting a tool he refers to as Fable 5 inside Claude Code with the following instruction:

Upgrade to anthropic>=1 - read https://raw.githubusercontent.com/anthropics/anthropic-sdk-python/refs/heads/main/MIGRATION.md and get the tests passing
- Simon Willison, August 24, 2026

The approach is notable for its directness: rather than summarising the migration requirements himself or writing the code changes by hand, Willison pointed the model at the upstream migration documentation and asked it to do the work end-to-end, with passing tests as the acceptance criterion. The resulting pull request was published and the release followed.

This workflow - using a live URL to a migration guide as context, then asking a code-capable model to execute the changes autonomously - reflects a pattern that has become increasingly common among developers maintaining open-source tooling. The cost of keeping a plugin current with upstream library changes falls substantially when a model can read the changelog, make the edits, and run the test suite without step-by-step human direction.

What this means for LLM plugin users

The llm-anthropic plugin provides access to Anthropic’s Claude series of models through Willison’s LLM tool, a command-line and Python interface for interacting with large language models. Users who keep their Python environments up to date will eventually pull in the anthropic v1.0.0 library automatically, at which point any version of llm-anthropic older than 0.27 would break. The 0.27 release closes that gap.

For developers who use the LLM ecosystem more broadly, the situation with llm-anthropic mirrors what anyone building on top of rapidly evolving AI SDKs faces routinely. Both Anthropic and OpenAI have now published major-version SDK releases within weeks of each other, each with the same fundamental transport-layer change. Plugin and integration authors who have not yet updated their code for the OpenAI v3.0.0 change are facing an equivalent task.

The broader pattern here is one of SDK consolidation. As Python’s HTTP client ecosystem matures and httpx2 becomes the preferred foundation, the AI SDK layer is standardising on top of it. That is a net benefit for long-term maintainability - a single modern transport dependency rather than a patchwork of older ones - but it compresses the maintenance window for anyone building on top of those SDKs. A major version bump in an upstream library can render a plugin non-functional overnight for users who update without checking compatibility.

Willison’s approach of using AI tooling to handle the migration itself points toward a practical response to that compression: if the migration guide is machine-readable and the codebase has a solid test suite, the cost of staying current can be delegated to the same category of tools that the plugin exists to support. The fact that the process worked cleanly enough to publish a release the same day the migration was prompted is a data point in favour of that approach - even if the test suite, not the model’s confidence, is ultimately what makes the result trustworthy.

// read moreLLM 0.33 Adds Repeatable Templates, OpenAI Library 3.x SupportAI ToolsLLM Command-Line Tool Gets Patch After OpenAI Library ChangeAI ToolsLLM-Gemini Plugin 0.33 Adds Gemini 3.7 Flash SupportAI Tools

Kinetyra