OpenAI released GPT-6.1 Sol on September 29, only a week after GPT-6 Sol. OpenAI says it “delivers near-Astra performance at a lower cost”, at a fifth of Astra’s list price, and some on Reddit even go so far as to call it Astra minor. Cost-effective Astra-level performance is great news. But GPT’s sweet spot is increasingly the lower bound, and OpenAI is no longer competing at the flagship level.

On Artificial Analysis’s Intelligence Index, GPT-6.1 Sol tops out at 51.8, just under Astra’s 52.7. Around the early 50s, it still holds its own against Claude Opus 5.5. At the same score, Sol costs a quarter to a half of what Opus does per task: at xhigh it lands next to Opus at medium, 51.0 against 51.2, for $0.39 against $1.34.

If you know your job will always need less than 52, that’s great. If you want anything above it, GPT simply doesn’t have a competitive one. On the Index, its only models past 52 are Astra at xhigh, 52.4 for $2.31, and Astra at max, 52.7 for $3.26. Opus 5.5 at high already scores 53.6 for $1.82, cheaper than either, and climbs to 57.6 at max, where GPT has nothing. Astra was supposed to be OpenAI’s Fable-class competitor, if not more. Opus, a tier below Fable, now beats both Fable and Astra at their top settings on score and price.

Most users would then ask why not keep only a Claude account and use Sonnet 5.5 instead of cheaper GPT models. At every score Sonnet 5.5 reaches, though, Sol or Opus gets there for less per task. Sonnet at max scores 56.0 for $7.60, and Opus at xhigh matches it for $3.46. Haiku 5.5 is expected next, likely aimed below Sonnet. If Anthropic manages, Claude will have the whole range covered, beating its competitors’ price point at every level.

From my own experience, the actual sweet spot is to have Opus 5.5 work on more complex scripts that compute locally, trimming off as much work as possible, and delegate the actual harder work to the agent. It is possible to have the script itself call for an agent, if the job gets too complicated; likewise, it is possible to have an agent bake that condition into the script itself. In this setup, GPT is quickly losing ground.