OpenAI announced GPT-6.1 Sol, an upgrade to GPT-6 Sol, positioning it as a lower-cost model that approaches the performance of its higher-tier GPT-6 Astra on several benchmarks. According to OpenAI, the model delivers improvements over GPT-6 Sol in coding, professional work, computer use, and scientific research, while priced at one-fifth of Astra's standard input and output token prices. Cached input is priced at $0.10 per million tokens, which OpenAI describes as 95% below standard input pricing.
OpenAI cited several benchmark results. On DeepSWE v1.1, a software-engineering evaluation, the company says GPT-6.1 Sol matches Astra at roughly one-fifth the cost and exceeds GPT-6 Sol's best score by 6.4 percentage points. On AutomationBench, which tests multi-step business workflows across 47 tools, OpenAI reports it scored 2.2 percentage points above Opus 5.5 at medium reasoning effort. On Terminal-Bench Science 0.1, OpenAI states it averaged $5.47 per task versus $23.21 for Opus 5.5 and $23.80 for Astra.
OpenAI also reports factuality gains, reducing the share of responses with a factual error from 11.4% to 7.7% at low reasoning effort on flagged prompts, and improvements on alignment evaluations. These figures are vendor-reported and were not independently verified.
- GPT-6.1 Sol is priced at one-fifth of GPT-6 Astra's standard input/output token prices
- Cached input costs $0.10 per million tokens, per OpenAI
- On Terminal-Bench Science 0.1, OpenAI reports $5.47 per task vs $23.21 for Opus 5.5
- Factual error rate dropped from 11.4% to 7.7% at low reasoning effort on flagged prompts
- GPT-6 Astra still scores highest on the hardest scientific tasks at 68.1%
What it means for you
OpenAI released a mid-tier model that does most of what its expensive top-tier model does, at roughly a fifth of the cost. In plain terms: if you or a developer already runs AI-powered tasks and the bill matters, there may now be a cheaper option that performs about as well. All the impressive numbers here come from OpenAI's own testing, so treat them as claims, not proven facts.
Try this
If you (or a developer you work with) are paying for automated AI tasks on a pricier model, run a small side-by-side test on GPT-6.1 Sol with your actual workload and compare cost and output quality before switching.
Who should care
Developers and technical teams running AI agents, code assistants, or document-processing pipelines at volume, where per-token cost affects the budget.
Skip this if
You use ChatGPT casually or through an app that picks the model for you — this pricing and benchmark news changes nothing about how you work.
Sources: OpenAI — read the original