Anthropic released Claude Fable 5.1, which scores highly on a new science benchmark.
In a creative test, its most expensive reasoning mode produced the best pelican image yet.
This mode cost $3.30 and took almost 14 minutes.
What happened
Anthropic released Claude Fable (and Mythos) 5.1 on September 1st, 2026. In a test, the model's 'max' reasoning setting produced the best AI-generated pelican image the author has seen from Anthropic, at a cost of $3.30 and 13 minutes 54 seconds of processing.
Why it matters
The model scored 52.6% on the new Terminal-Bench-Science 0.1 benchmark, up from 24.7% for Fable 5, 29.0% for Opus 5, and 22.4% for GPT-5.6 Sol. However, the author notes that at lower reasoning settings ('low' and 'medium'), the model appeared to skip reasoning entirely for the pelican prompt, costing around 10 cents.
What to watch
The 'max' reasoning level produced 65,927 output tokens and a detailed reasoning trace, leading to a pelican with a blue hat and a basket with a fish. When this output was animated with another prompt, it cost $1.37 and produced a video where the wheels rotated correctly in the original SVG.
Ask the AI about this article →
The release of Claude Fable 5.1 is highlighted by a significant jump on a brand-new scientific benchmark, Terminal-Bench-Science 0.1, where it scored 52.6% compared to 24.7% for its predecessor. This suggests a notable improvement in handling scientific research tasks, a key area for Anthropic's marketing.
However, the author's hands-on testing reveals a more nuanced story about the model's reasoning capabilities. The cost and time for generating a simple SVG varied dramatically with the reasoning effort setting, from about 10 cents and 24 seconds at 'low' to $3.30 and nearly 14 minutes at 'max'. Intriguingly, the lower efforts seemed to bypass reasoning entirely for this task.
The 'max' setting's lengthy reasoning trace shows the model engaging in detailed, deliberate planning, from checking visual collisions to fixing a bicycle fork's curve. This suggests that for complex, iterative tasks, the higher cost tiers provide a genuinely different and more capable mode of operation, though the author notes it still lacks the creative flair of a competing model like Gemini 3.7 Flash.
For example, today's edition would include:
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. Q&As are published on this page for other readers too.
Pasqal, four days after listing on Nasdaq, came to SEMICON Taiwan to argue that the industry's assumption that…
A senior Lam Research executive told an industry audience this week that the next major leap in semiconductor…
Nvidia invested $3.5 billion in MediaTek, a Taiwanese chipmaker, to help customers build custom AI chips that…
Dell issued fiscal year 2027 revenue guidance about $25 billion above analyst estimates, after sales surged 58…
Semafor and Riddance AI uncovered a network of about a dozen YouTube channels using real actors with AI-genera…
At a recent ICRA panel, robotics researchers discussed how to handle the overwhelming number of publications
