Claude Fable 5.2 在 VIALS-AA 科学视觉推理基准测试中领先

发生了什么

Claude Fable 5.2 在科学视觉推理基准测试 VIALS-AA 中领先；它调用工具的次数超过 Anthropic 的任何其他模型。与之不同的是，OpenAI 模型越智能，调用工具的次数越少 VIALS-AA 是我们对 @joinHandshakeAI 的视觉推理基准测试 VIALS 的实现，详见

接下来读什么

• 今日还有哪些重要变化Google DeepMind 推出 Gemini 4 Hydrogen，加速智能体工作流

这件事到目前为止

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2, lowering API prices to $8 per million input tokens and $40 per million output tokens, from $10 and $50. Anthropic says both models improve coding, reasoning, and vision, and that Fable produces responses faster than version 5.1. Fable is available to paid users with safeguards for risky cybersecurity and biology tasks; Mythos is limited to organizations approved through Anthropic’s verification programs. The launch reports cite strong benchmark results, including Fable’s first-place score on Artificial Analysis’ Intelligence Index v5 at its highest compute tier. Those figures come from the named benchmark evaluations; Anthropic’s prerelease examples also include software migrations, electronic devices, and physics problems. TechCrunch reports that an optional faster serving mode costs twice as much, while the lower API prices apply to standard use.