OpenAI released GPT-6.1 Sol on September 29, according to its API changelog. The company positions it as a lower-cost option for complex coding, computer use and professional work, approaching Astra’s performance. That is the maker’s assessment, not an independent finding that the two models are interchangeable.
The model page lists Standard API rates of US$2 per million input tokens and US$10 per million output tokens, with cached input at US$0.10. Its context window is 1,050,000 tokens and maximum output is 128,000 tokens. Tool calls use the Responses API; Chat Completions supports requests without tools. Comparing only the advertised input price misses output, tools and repeated attempts.
OpenAI’s product model guide describes a rollout in Work and Codex for Plus, Pro, Business, Enterprise and Edu. Enterprise and Edu need administrator enablement. Free and Go were excluded at launch, and Sol was not available in Chat. These distinctions matter: seeing a model in API documentation does not mean it appears in every ChatGPT conversation or account.
The model also supports beta multi-agent delegation through the Responses API. This lets developers divide work among cooperating agents. As an example, separate agents might inspect code and review a specification before a final answer is assembled. This is a possible workflow, not a guarantee that extra agents improve accuracy or reduce cost; coordination itself can consume time and tokens.
Standard and Fast launch first in the product rollout; Ultrafast for Sol is described as forthcoming. The model documentation supports US and EU data residency but excludes Fast mode with EU residency. Teams should check their actual processing configuration and account availability before choosing a workflow that depends on either speed or regional processing.
The useful question is whether Sol completes a real task reliably enough at its measured cost. Test it on representative examples, retain an approval step before sending or changing external information, and compare the complete result with the stronger model where errors are expensive. The facts here were checked against OpenAI’s changelog, model page and product guide on October 3, with September 29 retained as the release date.