llm-anthropic 0.29 put Opus 5.5 in the terminal on launch day

Simon Willison released llm-anthropic 0.29 on September 22. Its only listed change is support for Claude Opus 5.5, making the model available from a terminal through llm -m claude-opus-5.5. The GitHub release confirms the version, date, and model identifier.

Version 0.29 adds one important route to the model

The plugin does not improve the model itself. It maps Anthropic's interface into the LLM tool, where the same prompts, logging, and scripts can be used across providers. That thin integration layer determines whether a new API reaches everyday work on launch day.

Anthropic lists Opus 5.5 at $4 per million input tokens and $20 per million output tokens. The company also says typical workloads cost 40% less than Opus 5 and output is generated more than 30% faster.

A CLI shortens the path from comparison to automation

For developers, interchangeability is the useful part. The new model can be placed in an existing shell script, compared with other models on the same prompt, and logged without waiting for a new graphical interface. That is more useful than another benchmark table because cost and quality show up on a team's own tasks.

The release also demonstrates the value of small open source adapters. A vendor exposes an API, but community integration determines how quickly an announcement becomes a repeatable test.

A model identifier does not establish reliability

Version 0.29 adds support, not independent verification of Anthropic's claims. It says nothing about regressions, tool use compatibility, or whether Opus 5.5 will be cheaper on a particular task. In agent workflows, the number of steps and retries can matter as much as token prices.

Internal evals will reveal the value of fast integration

The release earns its keep when teams run the same task set through Opus 5 and Opus 5.5 and measure cost, time, and success rate. Plugin fixes involving streaming, tool use, and API changes will be another signal. They will show whether a one-day integration survives routine use.

Lilith's verdict

A model launch ends only when a developer runs it through a real test in the terminal. Version 0.29 opened the route, but every team still has to calculate its own Opus 5.5 bill.