GPT-6.1 Sol costs $2 and $10 per million tokens, and halves GPT-6 Sol's cache rate
OpenAI's new GPT-6.1 Sol lists at $2 per million input tokens and $10 per million output tokens on short-context requests, and its cached input rate of $0.10 per million is half the $0.20 GPT-6 Sol charges. Both rates come from OpenAI's API pricing page, where GPT-6 Sol stays at $2 input, $0.20 cached input and $10 output.
OpenAI's title card for its GPT-6 prompt caching changes, a separate announcement from GPT-6.1 Sol's list prices. Image: OpenAI Image: OpenAI
This is a separate model's list price, not the GPT-6 prompt caching discount and its 30-minute reuse window. GPT-6 Sol is still listed at its own rates.
GPT-6 Astra costs five times as much
GPT-6 Astra lists at $10 per million input tokens and $50 per million output tokens on short-context requests, so GPT-6.1 Sol undercuts it by a factor of five on both. OpenAI's own framing, in its announcement, is that GPT-6.1 Sol "nearly matches GPT-6 Astra's intelligence on agentic coding, computer use, and professional work at one-fifth of Astra's standard input and output token prices".
A long prompt costs double. Above 272,000 input tokens, GPT-6.1 Sol is billed at $4 input, $0.20 cached input and $15 output per million, because OpenAI charges 2x on input and cache rates and 1.5x on output for the whole request once a prompt crosses that line.
The benchmark figures all come from OpenAI
On Terminal-Bench Science 0.1, OpenAI says GPT-6.1 Sol at maximum reasoning effort costs $5.47 per task on average against $23.21 for Claude Opus 5.5 and $23.80 for GPT-6 Astra, while Astra keeps the highest score among the models tested at 68.1%. On AutomationBench the company puts GPT-6.1 Sol 2.2 percentage points above Opus 5.5 at medium reasoning effort, and 4.8 points above GPT-6 Sol at the same setting.
On DeepSWE v1.1 it claims the model beats GPT-6 Sol's best score by 6.4 percentage points at a lower reasoning effort and cost, and on OSWorld 2.0's offline set it comes within 2.1 points of GPT-6 Astra at maximum effort. OpenAI ran its own models for every one of those figures and took the competing models' numbers from published reports, and it warns that its factuality test uses deliberately difficult prompts: on those, at low reasoning effort, the share of answers containing a factual error falls from 11.4% for GPT-6 Sol to 7.7% for GPT-6.1 Sol.
Where the model turns up, and where it does not
GPT-6.1 Sol takes a 1,050,000-token context window with a maximum of 922,000 input tokens, returns up to 128,000 output tokens and carries a knowledge cutoff of April 30, 2026. It accepts text and images and returns text, supports the Responses, Chat Completions and Batch endpoints, and nothing else: no Realtime, no fine-tuning, no embeddings and no audio.
In ChatGPT it reaches Work and Codex, "starting with Pro users and expanding to Plus, Business, Enterprise, and Edu", according to OpenAI's release notes for September 29. OpenAI says the model is not yet available in Chat, and that a GPT-6.1 Sol Ultrafast with up to eight times faster token generation in Codex is coming "in the coming days".
OpenAI halved the API prices of GPT-6 Sol and Luna on September 25, four days before this model appeared at the same input and output rates as GPT-6 Sol. The one number that moved is the cache.
