Ultrafast is OpenAI’s new premium speed tier, launched at DevDay on September 29, 2026. In Codex it makes GPT-6 Astra generate tokens up to 8x faster, which OpenAI puts at 300 tokens per second. The speed isn’t cheap: at launch, the only personal plan with it is the new Pro 500, and it uses up your included usage at eight times the normal rate. Developers can also buy it in the API, at six times the standard price.
The short version
- GPT-6 Astra Ultrafast generates tokens up to 8x faster than Standard mode in Codex.
- In ChatGPT Work and Codex it’s on Pro 500 and eligible Enterprise and Edu plans; at launch, buying credits doesn’t unlock it on Pro 100 or Pro 200.
- It draws included usage at 8x the Standard rate, and purchased credits at 6x.
- In the API it costs $60 per million input tokens and $300 per million output tokens for short-context prompts, and runs only on US or global processing.
Where Ultrafast is available
Ultrafast works in ChatGPT Work and Codex, where you pick it in the model picker, and in the OpenAI API. In ChatGPT Work and Codex it’s GPT-6 Astra only for now; in the API, organizations that work with an OpenAI account team can also ask for preview access on GPT-5.6 Sol. OpenAI’s Thibault Sottiaux (Tibo) said it is “coming soon for 6.1 Sol”, and the GPT-6.1 Sol launch post says “in the coming days”.
Ultrafast availability by plan
| Plan | Ultrafast |
|---|---|
| Pro 500 | Included, then paid from credits |
| Enterprise and Edu | Eligible plans; off by default on Enterprise until an owner enables it |
| Pro 100, Pro 200 | Not available at launch, even with credits |
| Plus, Business, Go, Free | Not available at launch |
| OpenAI API | All API users, at low rate limits |
On Enterprise, the workspace must use a credit-based or USD usage-based agreement; older Enterprise plans billed on rate limits aren’t supported. Workspaces that require inference to stay in a region outside the US can’t use Ultrafast, according to OpenAI’s speed settings page.
How much of your allowance Ultrafast uses
Speed modes in ChatGPT Work and Codex use your included limits and credits at a higher rate than Standard mode on the same model. OpenAI stresses that these multipliers describe billing, not how much faster the model runs.
Usage rates for each speed mode
| Mode | Included usage | Credits and pay-as-you-go | Plans |
|---|---|---|---|
| Standard | 1x | 1x | All plans |
| Fast | 2.5x | 2x | Supported models, where available |
| GPT-6 Astra Ultrafast | 8x | 6x | Pro 500, eligible Enterprise and Edu |
Pro 500 includes 25 times the Plus allowance. Spent entirely in Ultrafast at 8x, that goes about as far as a bit over 3x the Plus allowance used at Standard speed. Once the included usage runs out, Ultrafast draws on your available credits. Pro 500 costs $500 a month, or ₹52,900 in India including GST; our Pro 100 vs Pro 200 vs Pro 500 comparison has its prices in the US, UK, Europe and India.
Ultrafast in the OpenAI API
Developers turn it on by setting service_tier to ultrafast with gpt-6-astra in the Responses API. OpenAI strongly recommends a persistent WebSocket connection for agents that make many tool calls in quick succession, because without one, network overhead can reduce the latency gains. Ultrafast also works over plain HTTP.
GPT-6 Astra API prices per million tokens
| Tier | Input | Cached input | Output |
|---|---|---|---|
| Standard | $10 | $1 | $50 |
| Fast | $20 | $2 | $100 |
| Ultrafast | $60 | $6 | $300 |
Long-context prompts cost $120 input and $450 output per million tokens in Ultrafast. The table shows short-context prices from OpenAI’s API pricing page. Default Ultrafast rate limits are 500,000 tokens per minute on usage tiers 1 to 3, 1 million on tier 4 and 5 million on tier 5. OpenAI’s Ultrafast guide says the API is “up to 8x faster” than Standard, while its DevDay recap puts the API speedup at up to 6x.
Data residency
Ultrafast supports US data residency and global processing only. EU and other regional processing endpoints aren’t supported, so teams that need processing kept in the EU or another non-US region can’t use it.
FAQs
Can I get Ultrafast on Pro 200 by buying credits?
No. OpenAI says buying credits on Pro 100 or Pro 200 doesn’t unlock Ultrafast at launch.
Is Ultrafast available for GPT-6.1 Sol?
Not yet. OpenAI says it’s coming. For now, Ultrafast is broadly available only for GPT-6 Astra.
Does Ultrafast make Astra smarter?
No. It’s the same model generating tokens faster, at a higher price.
Related
Leave a Reply