The GPT-6.1 Sol model delivers intelligence capabilities nearly identical to OpenAI's flagship Astra for roughly 20% of the standard API pricing. This new mid-range tool scores just one point below Astra in the Intelligence Index while costing less than a quarter per task at maximum effort. To support the deployment of long-running agents, OpenAI has set cached input costs at $0.10 per million tokens, providing a 95% discount relative to standard input fees.

The release replaces the original GPT-6 Sol only seven days after its launch, following reports that the previous version was an unreliable preview. While GPT-6.1 Sol is more cost-efficient—utilizing 40% to 67% fewer input tokens on most benchmarks—it operates 2 to 3 times slower than its predecessor on agentic tasks. The model is now generally available via the API and integrated into tools including GitHub Copilot.