OpenAI ships GPT-6.1 Sol at one-fifth Astra cost

OpenAI has upgraded GPT-6 Sol into GPT-6.1 Sol and is pitching it as a cost-balanced option rather than as its outright best model. The company says GPT-6 Sol already handled agentic coding, specialist tasks and computer operation well, closing in on GPT-6 Astra's performance on hard tasks while keeping standard input/output token pricing to one-fifth of Astra's. The new model pushes that idea further, and OpenAI is also launching GPT-6 Astra Ultrafast and GPT-6.1 Sol Ultrafast, which it says reach up to 8x the speed of GPT-6 Astra and are available via API as well as in ChatGPT Work and Codex for users on a new $500 monthly Pro tier.

The benchmark claims are framed almost entirely around cost. On DeepSWE v1.1, GPT-6.1 Sol is said to match GPT-6 Astra's performance at about one-fifth the cost, and to beat GPT-6 Sol's top score by 6.4 percentage points with less thinking and lower cost. On GDP.pdf it scores higher than Opus 5.5 at under half the per-task cost across the tested reasoning settings. On AutomationBench it beats Opus 5.5 by 2.2 points at roughly one-third the cost, and on OSWorld 2.0's offline set it comes within 2.1 percentage points of Astra while keeping per-task cost to about one-seventh.

That leaves the accuracy trade-off as the thing to weigh. Under the "very high" thinking setting, GPT-6.1 Sol's factual-error rate is 4.1%, down from GPT-6 Sol's 4.5% and close to Astra's 4.0% — still marginally behind the frontier model. OpenAI itself frames GPT-6 Astra as its top frontier model, so GPT-6.1 Sol looks aimed at teams that want to run important work more frequently, and at API users building and operating applications at scale, rather than at buyers who need the absolute best score at any price.