Meta Muse Wonder 评测：AA 52 分，与 Gemini 4 Argon 持平
Meta 新旗舰 Muse Wonder 在 Artificial Analysis 拿到 52 分，与 GPT-6 Astra、Gemini 4 Argon 同档，但落后 Claude Opus 5.5。
Meta shipped a frontier model that is better than Astra and Argon. here's my review. 1 → Muse Wonder - what Meta said it’s for personal and professional agents, memory across conversations, coding, law/finance/health work. 2 → better than the last Muse? yes. Artificial Analysis 52 at max vs Muse Spark 1.3 at 48. Terminal-Bench 57% vs 33%. 3 → rank in its field Artificial Analysis 52. level with GPT-6 Astra and Gemini 4 Argon (~53). behind Claude Opus 5.5 (57.6) and Sonnet 5.5 (56). 4 → strengths big jump on agentic coding (Terminal-Bench). multimodal (text, image, video, audio). 1M context. available now on the Meta Model API and Muse Code. 5 → weaknesses expensive for the score. $4.52 per AA task vs Sol’s $0.72. high input token use. still trails Opus and Sonnet on intelligence. Wonder Full (56) has no public API. 6 → models still above it Claude Opus 5.5 and Sonnet 5.5. side by side with Astra and Argon. 7 → price per 1M tokens $4 in / $20 out. cached input $0.20. same as Opus 5.5, but 2x Sol and Sonnet 5.5.