8 个模型按 Artificial Analysis 分数与单任务成本排名
Meta 首个前沿模型 Muse Wonder 排到并列第二，Claude Opus 5.5 居首；GPT-6 Astra 落在 B 档、每任务 3.26 美元，GPT-6.1 Sol 以 0.72 美元成性价比之选。
Meta's first frontier model just dropped. Its best version ties for 2nd out of 8: I ranked 8 models by Artificial Analysis' score. And I added what each costs to run one test task. New to AI? Start here: https://charliehills.substack.com/p/resource S-TIER (58/100) ↳ Claude Opus 5.5 - top score, $5.98 a task A-TIER (56/100) ↳ Claude Sonnet 5.5 - $5.46 a task ↳ Muse Wonder Full - new, Muse Code only B-TIER (53/100) ↳ GPT-6 Astra - $3.26 a task ↳ Claude Fable 5.1 - $7.63 a task ↳ Gemini 4 Argon - $1.99, limited access C-TIER (52/100) ↳ Muse Wonder - new, $4.52 a task ↳ GPT-6.1 Sol - $0.72, best value on the list So is Meta back? I'd say close, but behind Opus 5.5 (for now). You pay Opus 5.5 prices for Muse Wonder. And you get six points less. If value matters, I'd look at GPT-6.1 Sol. It scores 52 for $0.72 a task. Repost ♻️ to help someone in your network.
这条是引用着另一条写的：x.com/finkd/status/2109689305073614918