Anthropic released two new models, Fable 5.2 and Mythos 5.2, on Nov 9 local time. The same day, developer Simon Willison said his developer tool llm-anthropic had been updated to version 0.31, adding support for both models so users can call Fable 5.2 directly from the command line to generate content.
Tool update and how to call the models
According to Willison's release notes, users who install llm and the llm-anthropic plugin and configure an Anthropic API key can send prompts to the model with the llm -m claude-fable-5.2 command. He demonstrated the workflow by generating an SVG image of a pelican riding a bicycle.
Willison tested all five reasoning levels: low, medium, high, xhigh and max. He said xhigh produced the best results, while max returned no image at all — the model exhausted its token budget after spending 128,000 tokens thinking.
Willison noted this is the same flaw he observed on Opus 5.5, but a new problem for the Fable line: earlier Fable 5 and Fable 5.1 both returned images normally at the max level.
Community feedback: a new leader, and "limited gains"
On the V2EX community, one user posted that Fable "officially launched with 65 points and stays at the top of the leaderboard," adding that "everyone complains, but the model delivers — it's satisfying to run." The same post contrasted it with another vendor's new model (referred to in the post as "o/"), saying that company "doesn't dare release first." The post did not give the source of the 65 score or the benchmarks behind it.
In another V2EX thread, a user gave a different assessment of Fable 5.2 in practice: "Personally, I don't feel much of an improvement — maybe Opus 5.5 is already strong enough." That feedback contrasts with the "top of the leaderboard" claim, suggesting some developers are reserving judgment on the new model's marginal gains over its predecessor.
What to watch next
Community discussion is currently focused on the gap between Fable 5.2's benchmark ranking and real-world experience, and on whether the token-exhaustion problem at the max reasoning level will be fixed in a later version. The specific capabilities and benchmark data for Mythos 5.2 have not been disclosed in the available material and await further information.