Cached input at $0.10 per million tokens. That single number is probably the most important thing in OpenAI’s GPT-6.1 Sol release. It’s 95% cheaper than standard input pricing and undercuts the previous GPT-6 Sol cached rate by half. For developers building agents that reuse context across long sessions, that’s not a minor discount — it changes what’s economically viable to build.
OpenAI announced GPT-6.1 Sol as a direct upgrade to GPT-6 Sol, positioning it just below GPT-6 Astra in capability while pricing it at roughly one-fifth of Astra’s standard token rates. The standard API prices are $2 per million input tokens, $0.10 per million cached input tokens, and $10 per million output tokens. GPT-6 Astra remains the top of the stack for the hardest tasks, but 6.1 Sol is clearly designed to handle the 80% of agentic workloads that don’t need maximum intelligence.
What the benchmarks actually show
The benchmark story is reasonably strong, though context matters. On DeepSWE v1.1, a coding evaluation run on real codebases, GPT-6.1 Sol matches GPT-6 Astra’s score at about one-fifth of the cost and beats GPT-6 Sol by 6.4 percentage points at lower reasoning effort. On GDP.pdf, which tests professional document comprehension across finance, healthcare, and legal domains, it beats Anthropic’s Opus 5.5 with fallbacks at less than half the cost per task.
On AutomationBench, which runs agents through multi-step business workflows across 47 tools, GPT-6.1 Sol scores 2.2 points above Opus 5.5 at medium reasoning effort and at roughly a third of the cost. These are meaningful margins in a competitive field where Claude, Gemini, and GPT are all fighting over enterprise workflow budgets.
Computer use is another area where the model makes real progress. On OSWorld 2.0’s offline set, it outperforms GPT-6 Sol by seven percentage points at maximum reasoning effort and comes within 2.1 points of Astra at about one-seventh the cost. For teams building desktop or browser automation agents, that cost gap is significant.
Where it pulls ahead — and where it doesn’t
Scientific research benchmarks are where GPT-6.1 Sol’s cost advantage gets most dramatic. On Terminal-Bench Science 0.1, it averages $5.47 per task at maximum effort, compared to $23.21 for Opus 5.5 and $23.80 for Astra. Still, OpenAI is clear that Astra holds the highest score at 68.1% and should be used for the most demanding scientific work. GPT-6.1 Sol is not the right tool for every job.
Factuality also improves. At low reasoning effort, the rate of responses containing a factual error drops from 11.4% to 7.7% compared to GPT-6 Sol, a 32% reduction. The error rate stays within 1.9 percentage points of Astra’s across tested settings, which is a solid result for a model priced this much lower.
Safety and alignment improvements
OpenAI is also citing alignment gains. GPT-6.1 Sol fails to disclose a broken search tool in 2.1% of test cases, down from 4.9% for GPT-6 Sol. It shows no attempts to bypass an automated safety reviewer, matching both Astra and the earlier Sol. These are stress-test scenarios, not typical usage patterns, but the direction of improvement matters for anyone deploying autonomous agents in production.
Availability and what comes next
The model is available today for Plus, Pro, Business, Enterprise, and Edu users in ChatGPT Work and Codex, and accessible via the API as gpt-6.1-sol. It’s not yet available in standard Chat. OpenAI also says a GPT-6.1 Sol Ultrafast variant is coming within days, offering up to 8x faster token generation in Codex.
- Standard input: $2 per million tokens
- Cached input: $0.10 per million tokens
- Output: $10 per million tokens
- Available in ChatGPT Work, Codex, and the OpenAI API
- GPT-6.1 Sol Ultrafast coming soon with up to 8x speed in Codex
The bigger picture here is that OpenAI is compressing the capability-cost curve faster than most expected. Six months ago, Astra-level coding performance required Astra-level pricing. Now it doesn’t. That matters most to developer teams running high-volume agents where compute cost directly limits how ambitiously they can build. Anthropic and Google will respond, but right now GPT-6.1 Sol puts real pressure on the mid-tier of the market.
