OpenAI has released GPT-6.1 Sol, a model the company positions as coming close to GPT-6 Astra's agentic coding and computer-use performance while costing a fraction as much to run. Sol is priced at $2 per million input tokens and $10 per million output tokens, against Astra's $10 and $50, a fifth of the cost at both ends.
Cached input pricing drops further still, to $0.10 per million tokens, which OpenAI says is 95 per cent below the standard input rate and half of the previous GPT-6 Sol's cached rate. For workloads that repeatedly reuse large chunks of context, such as long coding sessions or document-heavy agents, that cache discount compounds the headline saving.
On performance, OpenAI reports Sol scoring 75.2 per cent on DeepSWE v1.1 at its highest reasoning effort, edging past Astra's 74.1 per cent on the same test. The picture is more mixed elsewhere: independent reporting puts Sol behind Astra by 2.1 points on the offline OSWorld computer-use benchmark, 5.3 points on AutomationBench, and 11.1 points on Terminal-Bench Science. So "near-Astra" is accurate on some tasks and a genuine step down on others, not a uniform match.
Sol is live now in ChatGPT Work and in Codex, and available through the API under the identifier gpt-6.1-sol. It has not yet rolled out to the standard ChatGPT consumer chat interface. As with any vendor-reported benchmark set, these figures come from OpenAI itself, and independent, third-party evaluation of Sol was not yet available at launch, so teams weighing the switch from Astra should run their own workload-specific comparisons before committing.
Source: MarkTechPost
