<p>Today is <a href="https://www.anthropic.com/claude-fable-and-mythos-5-1">Claude Fable (and Mythos) 5.1 day</a>. Anthropic say that Fable 5.1 "sets a new standard for coding, knowledge work, and long-running problem-solving tasks". Their announcement spends a notable amount of time on scientific research, boasting of a 52.6% score on the brand new <a href="https://www.terminal-bench-science.ai">Terminal-Bench-Science 0.1</a> benchmark (first announced <a href="https://www.tbench.ai/news/terminal-bench-science-0-1">on August 27th</a>), up from 24.7% for Fable 5, 29.0% for Opus 5 and 22.4% for GPT-5.6 Sol. Other benchmarks show slightly improved scores, but none as impressive as the Science one.</p>
<p>But how well can it pelican?</p>
<p>Back in July <a href="https://simonwillison.net/2026/Jul/16/kimi-k3/">I wrote about</a> how I was losing faith in the pelican benchmark - its connection to how good the models were at other tasks didn't seem to hold as strongly as it did <a href="https://simonwillison.net/2025/Jun/6/six-months-in-llms/">back in 2025</a>. The most interesting insights I get from it now are comparisons within model families, and particularly comparisons for the same prompt at different reasoning effort levels.</p>
<p>Fable 5.1 has five reasoning levels: low, medium, high, xhigh, max - and no option to turn off reasoning entirely.</p>
<p>I fixed <a href="https://github.com/simonw/llm-anthropic/issues/88">an issue</a> in <a href="https://github.com/simonw/llm-anβ¦
https://simonwillison.net/2026/Sep/1/claude-fable-5-1/
this is a test share