GPT-6.1 Sol Ultrafast, OpenAI’s new speed tier for its mid-priced model, costs $12 per million input tokens and $60 per million output tokens in the API. That is six times what Sol costs at standard speed, and 20% more than OpenAI charges for its flagship GPT-6 Astra at standard speed ($10 and $50).
OpenAI launched Sol on September 29 as “near-Astra intelligence for a fifth of the price.” Ultrafast, which went live on October 8, is the upgrade for people who would rather wait less than save money. It matters now because OpenAI is promoting it as “up to 8x faster,” while its own price list, its help center and one small independent test tell a more complicated story.
GPT-6.1 Sol Ultrafast costs more than Astra at standard speed
OpenAI’s API price list, which we checked on October 11, puts the speed tiers side by side. Every Sol tier multiplies the base price: Fast is twice Standard, Ultrafast six times.
| Speed tier (per 1M tokens) | GPT-6.1 Sol, input / output | GPT-6 Astra, input / output |
|---|---|---|
| Standard | $2 / $10 | $10 / $50 |
| Fast | $4 / $20 | $20 / $100 |
| Ultrafast | $12 / $60 | $60 / $300 |
Two things can push the bill higher. Prompts over 272,000 input tokens are billed at long-context rates for the whole request, which for Sol Ultrafast means $24 input and $90 output per million. Regional processing, the US or EU data-residency option, adds 10%. As a rough example, our arithmetic from OpenAI’s rates: a 20,000-token prompt with a 3,000-token answer costs about $0.07 at standard speed and $0.42 on Ultrafast. It’s the same model ID, gpt-6.1-sol, with service_tier: "ultrafast" added to the request, and six times the bill.
“Up to 8x faster” is OpenAI’s number, and it has fine print
OpenAI’s October 8 post on X promises “near-Astra intelligence at up to 8x faster speeds than Sol Standard.” The price graphic it posted a few seconds later labels that speed row “API token generation” with “within-model baselines,” which reads as each model being measured against its own Standard speed. The Ultrafast guide and the Sol model page give no tokens-per-second figure for Sol, and OpenAI hasn’t said how Sol Ultrafast’s speed compares with Astra’s.
When Ultrafast launched for Astra on September 29, OpenAI’s developer account said the tier offers “up to 8x faster token generation (300 tokens per second) in Codex and up to 6x in the API.” We couldn’t find a Sol-specific API multiple on OpenAI’s pages. Faster token generation also isn’t the same as a finished job: tool calls, test runs and network waits don’t speed up.
One independent test lands lower. The AI-tools site Kingy AI ran four matched pairs per task on gpt-6.1-sol at medium reasoning effort and reported median completion times of 9.61 seconds on Standard against 2.61 on Ultrafast for a coding fix (about 3.7 times faster) and 16.14 against 4.17 seconds for a research brief (about 3.9 times). Visible output ran at 78.7 against 278.7 tokens per second on the coding task, about 3.5 times, and 65.2 against 290.9 on the brief, about 4.5 times. Those are small, workload-specific, client-side numbers rather than a benchmark, and Kingy says so itself.
Most ChatGPT plans can’t turn it on
The API is open. OpenAI’s Ultrafast guide says the mode for Sol is available to all API users, and the changelog adds “subject to rate limits.” ChatGPT is stricter. According to OpenAI’s help center, Ultrafast in Work and Codex is available at launch on the $500 Pro plan and in eligible Enterprise and Edu workspaces. It is “not available at launch on Plus, Pro $100, Pro $200, or Business, including through credits.”
For Pro $500 subscribers, the help page says Ultrafast uses included Work and Codex usage at 8 times the standard rate for the model you pick, then draws credits at 6 times. Enterprise and Edu workspaces pay from workspace credits, and access depends on workspace permissions.
What OpenAI hasn’t shown yet
Ultrafast is a service tier, not a new model, so the quality question is whether speed costs anything. We found no benchmark that tests the Ultrafast tier separately. OpenAI’s Sol scores come from its September 29 launch post, which says the company ran its own models in its research environment or through its API, and the “near-Astra” claim rests on those numbers. OpenAI’s guide says only to use Ultrafast “when speed justifies the higher cost.”
Two things to watch. One is whether OpenAI opens Ultrafast to Plus and the $100 and $200 Pro plans, which the help page frames as a limit “at launch.” The other is whether it publishes a Sol-specific speed number for the API, so buyers can weigh a 6x price against a measured gain. Until then, the fair test is cost per finished task on your own workload, not cost per token.
Frequently asked questions
How much does GPT-6.1 Sol Ultrafast cost?
In the API it costs $12 per million input tokens and $60 per million output tokens for prompts up to 272,000 input tokens, which is six times standard Sol. Cached input is $0.60 per million. Above 272,000 input tokens the rates are $24 for input and $90 for output.
Can ChatGPT Plus use GPT-6.1 Sol Ultrafast?
No. OpenAI’s help center says Ultrafast isn’t available at launch on Plus, Pro $100, Pro $200 or Business, even with credits. It is offered on the $500 Pro plan and in eligible Enterprise and Edu workspaces.
How much faster is GPT-6.1 Sol Ultrafast?
OpenAI says up to 8x faster token generation than standard Sol. A small independent test by Kingy AI measured about 3.5x to 4.5x in output rate and 3.7x to 3.9x in completion time on two tasks, so your results will depend on the workload.
How do I use GPT-6.1 Sol Ultrafast in the API?
Call gpt-6.1-sol in the Responses API with service_tier set to “ultrafast.” OpenAI recommends a persistent WebSocket connection for agents that make many tool calls, because network overhead can eat into the speed gain.
Sources
All accessed October 11, 2026.
- OpenAI, “Introducing GPT-6.1 Sol” (Sep 29, 2026): openai.com
- OpenAI API pricing, GPT-6 Astra, GPT-6.1 Sol and Ultrafast tables (checked Oct 11, 2026): developers.openai.com
- OpenAI API docs, GPT-6.1 Sol model page (price multipliers, context window, long-context rule): developers.openai.com
- OpenAI API docs, Ultrafast mode guide (availability,
service_tier, WebSocket advice): developers.openai.com - OpenAI API changelog, Oct 8, 2026 entry “Ultrafast mode for GPT-6.1 Sol”: developers.openai.com
- OpenAI Help Center, “ChatGPT Work and Codex” (Ultrafast plan rules, checked Oct 11, 2026): help.openai.com
- OpenAI Developers on X, Ultrafast for GPT-6.1 Sol (Oct 8, 2026, 2:26 PM ET) and the API price graphic: post, price graphic
- OpenAI Developers on X, “This is Ultrafast” (Sep 29, 2026): x.com
- Kingy AI, “GPT-6.1 Sol Ultrafast: Is the Time Saved Worth the Price?” (Oct 8, 2026; small matched trials by a third party): kingy.ai
Related: GPT-6 is free in ChatGPT, but free users get Luna, not Sol · Claude Haiku 5.5 matches GPT-6 Luna’s 10-cent price · More LLM news
Last updated: October 11, 2026. To report an error, see our corrections page. Articles are drafted with AI assistance and reviewed and edited by an editor; see our editorial policy.
