GPT-6.2 Astra 编程能力跃升但聊天能力下滑
原标题:GPT-6.2 Astra (Max) by @OpenAI ranks #6 on the Agent Arena Index, one place above GPT-6 Astra (Max).
58AI 研判分
Arena.ai 的 Agent Arena Index 显示,OpenAI 的 GPT-6.2 Astra (Max) 综合排名第 6,仅比 GPT-6 Astra (Max) 高一位。分项来看,新模型编程得分从 1,288 升至 1,338,排名由第 8 升至第 3;但聊天得分从 1,265 降至 1,211,排名由第 8 跌至第 11,两项变化在总分上几乎相互抵消。
报道正文
Arena.ai (@arena) · 2026-11-21 17
GPT-6.2 Astra (Max) by @OpenAI ranks #6 on the Agent Arena Index, one place above GPT-6 Astra (Max). However, these models excel in very different capabilities. Let’s dive in by index.
GPT-6.2 Astra vs GPT-6 Astra across key Agent Arena Indexes:
- Coding: 1,338 vs 1,288 (+50)
- Work: 1,280 vs 1,267 (+13)
- Chat: 1,211 vs 1,265 (-54)
This shows the new model’s gain is in coding, where GPT-6.2 Astra climbs from #8 to #3. On Chat, the everyday tasks such as research, tutoring or writing, it drops from #8 to #11. The two shifts nearly cancel out in the overall score.
