Simon Willison 对 Claude Fable 5 的实测：一个庞大、受限且昂贵的前沿模型
英文摘要
Anthropic released Claude Fable 5 alongside the unguarded Claude Mythos 5, both with 1M context, 128K max output, and Jan 2026 knowledge cutoff. Priced at $10/M input and $50/M output tokens, it is twice the cost of Opus 4.8 and notably slow but performed impressively on code-heavy tasks. Simon Willison found it spontaneously re-implemented his micropython-wasm project with full CPython, and via Claude Code it autonomously added human-in-the-loop tool approval to LLM and Datasette Agent, producing LLM 0.32a3. Its factual recall of niche open-source details suggests a very large model size. Daily usage cost reached $110.42 under a $100/month plan.
中文摘要
Anthropic 发布了 Claude Fable 5 和移除安全护栏的 Claude Mythos 5，两者均支持 100 万 token 上下文、12.8 万最大输出 token，知识截止 2026 年 1 月。定价为输入 $10/百万 token、输出 $50/百万 token，是 Opus 4.8 的两倍，速度较慢但代码能力突出。Simon Willison 用它成功将 micropython-wasm 升级为完整 CPython，并通过 Claude Code 为 LLM 和 Datasette Agent 添加了人工审批流程，最终推出 LLM 0.32a3。它能精确回忆大量小众开源项目细节，暗示模型规模极大。在 $100/月套餐内，当日花费达 $110.42。
关键要点
Claude Fable 5 and Mythos 5 launched simultaneously; Fable has strict safety guardrails and API fallback for rejected prompts, while Mythos lacks classifiers.
Claude Fable 5 与 Mythos 5 同时发布；Fable 有严格安全护栏和拒绝提示时的 API 回退机制，Mythos 则没有分类器。
Models feature 1M context, 128K max output, Jan 2026 knowledge cutoff, and twice the pricing of Opus 4.8 ($10/$50 per million input/output tokens).
模型拥有 100 万 token 上下文、12.8 万最大输出、2026 年 1 月知识截止，价格是 Opus 4.8 的两倍（输入/输出每百万 token 10/50 美元）。
Fable demonstrated deep factual knowledge, accurately listing many niche Simon Willison projects where Opus 4.8 was cautious, suggesting a very large parameter count.
Fable 展现出深度事实知识，准确列出大量 Simon Willison 的小众项目，而 Opus 4.8 较为保守，暗示参数规模极大。
In Claude.ai’s code environment, Fable autonomously upgraded a MicroPython sandbox to full CPython and produced a 13.9 MB wheel, solving complex build issues.
在 Claude.ai 代码环境中，Fable 自主将 MicroPython 沙盒升级为完整 CPython，生成 13.9 MB 的 wheel 包并解决构建难题。
Using Claude Code, Fable added human-in-the-loop tool approval to Datasette Agent and resolved four underlying issues in the LLM library, resulting in LLM 0.32a3 with refined tool APIs.
通过 Claude Code，Fable 为 Datasette Agent 添加了人工审批功能，并修复 LLM 库四个底层问题，催生了 LLM 0.32a3 及精细化的工具 API。
Daily token spend reached $110.42 within a $100/month subscription, highlighting the model's high operational cost for heavy use.
在 $100/月套餐内，当日 token 花费达 $110.42，凸显高强度使用下的高昂成本。