Initial impressions of Claude Fable 5

Open original

Simon Willison runs Claude Fable 5 through its paces, finding it slower and pricier than Opus 4.8 but far more knowledgeable thanks to its 1 million-token context. He tests it on real-world coding tasks—upgrading a MicroPython sandbox to full CPython in WASM and adding pause-resume hooks to Datasette Agent—showing it can build complex features end-to-end.

  • Fable 5 costs twice as much as Opus 4.5–4.8 ($10/$50 per million tokens vs their pricing) and runs slower, but handles every complex prompt thrown at it.
  • Its 1M-token context gives it dramatically deeper recall than Opus 4.8—listing a dozen detailed open-source projects with dates versus Opus's brief list of four.
  • It successfully converted a MicroPython-wasm project to full CPython in WASM, ultimately producing a working 13.9 MB wheel runnable via uv, demonstrating real end-to-end coding capability.

Claude Fable 5 arrives with a 1 million-token context window, a 128 000-token output cap and a knowledge cutoff of January 2026. Anthropic says it matches Claude Mythos 5 on raw performance but layers on strict safety classifiers; if a request trips a guardrail you’ll get a rejection notice or can auto-fallback to another model. At $10 per million input tokens and $50 per million output tokens, it’s twice the cost of Opus 4.5–4.8. It runs noticeably slower than Opus, but so far it’s handled every complex prompt thrown at it. Simon Willison ran a side-by-side on a prompt asking for all his open-source projects. Opus 4.8 gave a brief, cautious list of four main items. Fable 5 produced a dozen, complete with approximate release dates, from files-to-prompt (April 2024) and datasette-extract (2024) back through Django’s July 2005 open-sourcing. That depth of recall hints at a much larger parameter count, though Anthropic hasn’t published any model-size specs. In the Claude.ai environment—where every chat runs in a full container sandbox—you can install packages, clone GitHub repos and execute code. Willison tested Fable by converting his micropython-wasm project to full CPython in WASM. Fable identified Brett Cannon’s CPython-WASI builds, struggled with zip-based stdlib boots, then produced a Proof-of-Concept using a directory-preopen approach. After a few prompt tweaks it even generated a 13.9 MB wheel that you can run via “uv run --with <wheel-url> python -c…”, proving it can handle real-world coding workflows.

Ask about this article