0.30

• New model: Claude Sonnet 5.5: llm -m claude-sonnet-5.5 hi #96
• New llm anthropic refresh command, which fetches the models available to your API key from the Anthropic models API and caches them in anthropic_models.json in your LLM user directory. Models this plugin does not know about are registered using the capabilities the API reports for them: image and PDF input, thinking, effort, structured outputs and maximum output tokens. #77, #97
• New llm anthropic models command listing the models available to your API key. Add --json to see the full API response, including each model's capabilities. #77
• New llm anthropic count command, which uses the token counting API to count the input tokens for a prompt without running it. It accepts the same options as llm prompt, including system prompts, attachments, fragments, templates, tools, schemas, model options and -c to count a follow-up in an existing conversation. From Python, model.count_tokens() accepts the same arguments as model.prompt() and returns an integer. #92
• Fixed a bug where every follow-up prompt in a conversation after a refusal failed with a 400 error: "all messages must have non-empty content except for the optional final assistant message". #37, #94 - thanks Raj Royal