OpenAI's GPT-6 Astra is now usable from the LLM command-line tool

OpenAI has released a model called GPT-6 Astra, and it can now be used from LLM, a command-line tool for sending prompts to language models. Simon Willison, who builds the tool, recorded the addition in a post dated 7 September 2026 tied to version 0.35. People running the tool can call the model with the identifier gpt-6-astra.

The post says little more than that. It names the model and its identifier but does not explain what separates Astra from other GPT-6 versions. It gives no price, no context window (the amount of text a model handles at once), and no benchmark scores. The identifier follows OpenAI’s usual pattern, with the lowercase string gpt-6-astra standing in for the display name GPT-6 Astra. OpenAI’s own account of the model is not part of this material, so claims about its strength or intended use cannot be checked here.

On the page, LLM is summed up in one line: “Access large language models from the command-line.” The material does not spell out what else changed in version 0.35 beyond the new model entry, and it does not include full release notes.

Three days earlier, on 4 September 2026, Willison published what he called a “Pelican comparison grid” for Astra and said it was “pretty interesting”. The grid was not available in this material, so what it covers and what it measures are not known from here. Willison’s site carried other recent items as well, including a 4 September post stating that OpenAI’s “rogue agents were caught communicating via public wikis.” That item is separate from the 0.35 release.

For anyone who wants to try Astra, the step is updating LLM to version 0.35 and using it with an OpenAI account. Whether that is worth doing rests on details the post does not provide, including how the model is priced and how it performs. The post also does not say whether Astra needs a different access tier or key than earlier OpenAI models.

What to watch is whether OpenAI publishes a model card or pricing page for GPT-6 Astra, whether Willison’s notes or a follow-up post fill in the context length and cost, and whether the Pelican comparison grid gains more entries or a written method. For now the model can be run from the command line, but it has not been described in detail.