爆料称 Astra 6.2 表现超预期,或成全球最强模型
有爆料称 OpenAI 对 Astra 6.2 的强化学习调优效果超出预期,企业合作方预测它将成为全球最强模型。
Anthropic and OpenAI are experiencing exponential acceleration in model development. Early access partners are claiming that Fable 5.5 is an unusually massive leap above Fable 5.1 We already saw a similar leap from Opus 5 to Opus 5.5 and clearly Anthropic has had a breakthrough. OpenAI has always thrived in RL and the refinement of Astra 6.2 has worked even better than expected. Enterprise partners are predicting that Astra 6.2 will become the undisputed best model in the world. Google has been claiming RSI lately, but it is becoming clear that OpenAI and Anthropic are actually the labs getting close to removing the human from the loop.