用户吐槽 Opus 5.5 遇难题直接罢工,竞品却肯硬啃
imjustnewatai · x · 2026-11-18
A user reports that Opus 5.5 tends to bail on extremely hard problems, saying "bro this is way too difficult," whereas competing models like Fable, Astra, and Sol will attempt them anyway. The post calls on Anthropic to fix this behavior, highlighting differing trade-offs between honest refusal and confident attempts.
「模型」频道最新
- 澳大利亚夫妻团队用不足千万美元造出 220B 主权模型 — aakashgupta · 2026-11-18
- Gemini Decisions API 单次前向 150-250ms 完成多模态判定 — McDonaghMatthew · 2026-11-18
- OpenAI 向 Pro 200 用户开放 GPT-6.2 Sol「Ultrafast」极速档 — shaunralston · 2026-11-18
- 实测称 Astra 6.2 令人失望:写作倒退、编码常过早收工 — danshipper · 2026-11-18
- 开发者称每个新模型发布都成了 Opus 5.5 的广告 — dejavucoder · 2026-11-18
- 博主实测 GPT-6.2 Sol Ultra Fast:又快又聪明 — vista8 · 2026-11-18