GPT-6.1 Sol: OpenAI's New Model Nearly Matches GPT-6 Astra at One-Fifth the Token Price
OpenAI has released GPT-6.1 Sol, a model that nearly matches its flagship GPT-6 Astra on agentic coding, computer use, and professional work, while costing one-fifth of GPT-6 Astra's standard input and output token prices. That combination changes the math for anyone running frequent, high-volume model calls rather than occasional ones.
OpenAI has released GPT-6.1 Sol, a model that nearly matches its flagship GPT-6 Astra on agentic coding, computer use, and professional work, while costing one-fifth of GPT-6 Astra's standard input and output token prices. That combination changes the math for anyone running frequent, high-volume model calls rather than occasional ones.
A price cut that doesn't ask you to give up much
The headline fact is simple: GPT-6.1 Sol costs one-fifth of what GPT-6 Astra charges for standard input and output tokens, and cached input drops all the way to $0.10 per million tokens. What makes that number interesting rather than just cheap is what you get for it. On DeepSWE v1.1, GPT-6.1 Sol matches GPT-6 Astra's score at roughly one-fifth the cost. On GDP.pdf, it approaches GPT-6 Astra's performance at roughly one-fifth the cost per task, and it scores higher than Opus 5.5 with fallbacks on that same task at less than half the cost per task. None of this means the two OpenAI models are interchangeable across every workload, but it does mean the price gap is no longer proportional to the capability gap the way it once was.
Where the gains over GPT-6 Sol actually show up
GPT-6.1 Sol isn't just a discount version of GPT-6 Astra — it's also a clear step up from its immediate predecessor, GPT-6 Sol. It beats GPT-6 Sol's best DeepSWE v1.1 score by 6.4 percentage points. On AutomationBench at medium reasoning effort, it scores 4.8 percentage points above GPT-6 Sol at the same setting, and 2.2 points above Opus 5.5. On OSWorld 2.0's offline set at maximum reasoning effort, it outperforms GPT-6 Sol by seven percentage points while coming within 2.1 points of GPT-6 Astra. The largest jump shows up on Terminal-Bench Science 0.1 at maximum effort, where GPT-6.1 Sol more than doubles GPT-6 Sol's score, averaging $5.47 per task along the way. For context, GPT-6 Astra still holds the highest tested score on that benchmark, at 68.1%, so the ceiling hasn't moved — but the distance to it has shrunk considerably at a fraction of the cost.
Fewer wrong answers, and no attempts to route around oversight
Capability and cost are only half the story if the model gets careless along the way. GPT-6.1 Sol cuts its low-effort factual-error rate from 11.4% to 7.7% compared with GPT-6 Sol, and across tested reasoning settings its factual-error rate stays within 1.9 percentage points of GPT-6 Astra's. On alignment evaluations covering transparency, restriction-following, and authorization behaviors, GPT-6.1 Sol posts lower failure rates than GPT-6 Sol. OpenAI also reports that no attempts by GPT-6.1 Sol to bypass an automated safety reviewer were observed during testing. These are specific, bounded results rather than a general verdict on trustworthiness, but they matter for anyone deciding how much autonomy to hand a model that's about to get cheaper and more widely deployed.
Where you can actually put it to work
GPT-6.1 Sol is live now in ChatGPT Work and in Codex for Plus, Pro, Business, Enterprise, and Edu users, though it hasn't reached the plain Chat surface yet. Developers can call it directly through the OpenAI API using the identifier gpt-6.1-sol, which makes it straightforward to swap into an existing pipeline built around token-metered pricing. OpenAI has also flagged a GPT-6.1 Sol Ultrafast variant in the works, aimed at Codex, promising up to eight times faster token generation — a sign that this release is meant to be a working tier, not a one-off benchmark exercise. For teams evaluating whether to build agent workflows around a model like this, that's the practical starting point: check the benchmark deltas against your own task type, price out the API calls against $2 input and $10 output per million tokens, and decide from there whether the gap to GPT-6 Astra is one you can live with.
Specs
- Standard input price: $2 per million tokens
- Cached input price: $0.10 per million tokens
- Standard output price: $10 per million tokens
- API identifier: gpt-6.1-sol
- Planned Ultrafast speed improvement: Up to 8x faster token generation
Sources
- OpenAI — OpenAI