OpenAI has released GPT-6.1 Sol, a model aimed at multi-step AI agent work including software development, complex document analysis and computer use. The company positions Sol as a lower-cost alternative that approaches GPT-6 Astra in several internal evaluations, while remaining an upgrade over GPT-6 Sol.
In OpenAI’s DeepSWE v1.1 software-engineering benchmark, GPT-6.1 Sol matched Astra at roughly one-fifth of its cost, according to the company. Sol also exceeded GPT-6 Sol by 6.4 percentage points while using a lower reasoning setting.
OpenAI reported a 4.8-percentage-point improvement over the previous Sol model at medium reasoning on AutomationBench, which measures whether business workflows are completed correctly. On the offline OSWorld 2.0 computer-use benchmark, the gain reached seven percentage points, with less than half the cost per task. These are vendor-published benchmark results, rather than guarantees of performance on every deployment.
The company also reported stronger scientific-research performance. At maximum reasoning, GPT-6.1 Sol doubled GPT-6 Sol’s score on Terminal-Bench Science 0.1, although Astra retained the top score and OpenAI’s recommendation for the most difficult scientific work. OpenAI put Sol’s average cost per task at US$5.47, versus US$23.80 for Astra.
On a factual-accuracy test, OpenAI said the share of responses containing at least one error fell from 11.4% to 7.7% at low reasoning. The company cautioned that the questions were intentionally difficult and drawn from conversations in which users had reported mistakes, so the figures are not intended to represent typical usage.
Access and API pricing
GPT-6.1 Sol is available in ChatGPT Work and Codex, but not yet in Chat mode. Access is available to the following subscriber groups:
- ChatGPT Plus users
- ChatGPT Pro users
- ChatGPT Business users
- ChatGPT Enterprise users
- ChatGPT Edu users
Developers can access the model through the API as “gpt-6.1-sol.” Standard pricing is US$2 per million input tokens, US$0.10 per million cached input tokens and US$10 per million generated tokens. The cached-input rate is 95% below the standard input rate, a potentially significant difference for agents that repeatedly reuse the same context.
OpenAI also says a Sol Ultrafast variant is due to arrive in Codex in the coming days, with token generation claimed to be up to eight times faster.
