In benchmark tests covering scientific and coding agent scenarios, GPT-6.1-Sol generally performs slightly below GPT-6 Astra and Opus 5.5, though it outperforms both in a few specific tests. Moreover, its testing costs are roughly one-fifth of Astra's and significantly lower than those of Opus 5.5. If the model's real-world performance matches these benchmark results, its combination of low cost and extremely high output speed would make it an outstanding model.

Clearly, this is the true GPT-6-Sol; what was released previously could essentially be called GPT-6-Terra.Otherwise, the massive improvement seen in just one week would be inexplicable; if OpenAI could improve models at that rate, AGI would have been achieved long ago.