The llm-anthropic package version 0.26 introduces several new models and adds support for server-side tools, and requires the llm library version 0.32.
New models
This release adds three Claude 5 models: claude-fable-5, claude-sonnet-5, and claude-opus-5. These are variants in Anthropic's Claude 5 family.
Server-side tools and CLI changes
The release (issues #75, #76) adds server-side tools for WebSearch, WebFetch, CodeExecution, and AnthropicMCP. These tools are available via the llm CLI -T interface or in Python via the tools= argument. The previous -o web_search* options have been removed in favor of -T WebSearch.
Requirement: llm 0.32 and evented streaming
llm-anthropic 0.26 now requires llm>=0.32 (issue #79). Reasoning, tool calls, tool results, and server-side tool results now stream as typed events. Reasoning output for llm CLI prompts is displayed to standard error by default unless --hide-reasoning / -R is passed.
Simplified 'thinking' settings and defaults
Extended thinking options were simplified to two settings: thinking and thinking_effort, where thinking_effort accepts low, medium, high, xhigh, or max. Claude 5 models perform thinking by default. For Sonnet 5 and Opus 5, thinking can be disabled with -o thinking 0, while Fable 5 always performs thinking. The -R/--hide-reasoning flag now omits reasoning from responses and logs. The thinking_budget, thinking_display, and thinking_adaptive options have been removed.
Metadata
Tags: llm, anthropic, claude, model-context-protocol
The release references issues #75, #76, #79 and #80 and contains changes affecting documentation and CLI behavior. Users should note the llm 0.32 compatibility requirement and the new -T based tool integrations.
