On 17 November 2026, OpenAI opened Ultrafast mode to Pro $200 subscribers for the first time, starting with GPT-6.2 Sol. Pro $500 users now get Ultrafast with both GPT-6.2 Sol and GPT-6.2 Astra in ChatGPT Work and Codex.
Ultrafast makes the same model write its answer much faster. The catch is simple: it burns through your plan allowance much faster too.
This guide covers what Ultrafast is, who gets it, how to switch it on in every app, what it costs, and when it is worth it.
- What it is: the fastest speed setting in ChatGPT Work, Codex and the OpenAI API. Same model, same quality, faster tokens.
- Speed: up to 4.5x Standard speed on GPT-6.2 Astra and GPT-6.2 Sol.
- Cost: uses 6x your included plan usage on GPT-6.2 models. In the API it costs 3x Standard until 28 February 2027, then 6x.
- Plans: Pro $200 (Sol only), Pro $500 (Sol and Astra), and eligible Enterprise and Edu workspaces. Not Plus, Go, Free or Business.
- Turn it on: lightning bolt in the composer, or
/ultrafastin the Codex CLI.
What Is Ultrafast Mode?
Ultrafast is a speed setting, not a new model. You pick a model like GPT-6.2 Sol, then pick how fast it should run. OpenAI calls Ultrafast “the fastest available responses for latency-sensitive work.”
The model’s brain stays the same. Your reasoning effort setting stays the same. Only the token generation speed changes, and so does the usage you pay for it.
OpenAI now offers three speeds across ChatGPT Work, Codex and the API:
- Standard: the default speed and the cheapest per task.
- Fast: about 1.5x speed for more usage. This is the old “Priority processing”, renamed Fast mode on 30 July 2026.
- Ultrafast: the top tier, up to 4.5x speed on GPT-6.2 models, at a much higher usage cost.
A short history
Ultrafast is not brand new. OpenAI announced it at DevDay on 29 September 2026 alongside GPT-6.1 Sol, saying it would arrive in Codex in the following days. At that point it was a Pro $500 and enterprise perk.
The 17 November update brought GPT-6.2 Astra and Sol to ChatGPT Work and Codex. It also pushed Ultrafast down to the Pro $200 plan for the first time, but only with Sol. If you want the full launch story, read our GPT-6.2 Astra and Sol guide.
How Fast Is It? Standard vs Fast vs Ultrafast
OpenAI gives speed as a multiple of Standard on the same model. It does not publish one fixed tokens-per-second figure for the GPT-6.2 models.
| Speed | Models | Speed vs Standard |
|---|---|---|
| Standard | All models | 1x (baseline) |
| Fast | GPT-6.2 Astra and Sol, GPT-6.1 Sol and Terra, GPT-6 Astra, Sol and Luna | About 1.5x |
| Ultrafast | GPT-6.2 Astra, GPT-6.2 Sol | Up to 4.5x |
| Ultrafast | GPT-6 Astra (in Codex) | Up to 8x |
| Ultrafast | GPT-6.1 Sol | Up to 8x in Codex (DevDay claim) |
At DevDay, press reports quoted OpenAI’s recap at around 300 tokens per second for Ultrafast in Codex. Treat that as a launch figure for the older models, not a promise for GPT-6.2.
Note: “Up to” matters here. Speed depends on load, prompt length and tool calls. An agent that waits on tests or file reads will not finish 4.5x sooner, because only the writing part gets faster.
Which Plans Get Ultrafast
Ultrafast is a premium feature. Here is how access breaks down after the 17 November update.
| Plan | Price | Fast mode | Ultrafast |
|---|---|---|---|
| Free / Go | $0 / $8 per month | No (Standard only) | No |
| Plus | $20 per month | Yes | No |
| Business | $20-25 per user per month | Yes | No |
| Pro $100 | $100 per month | Yes | No |
| Pro $200 | $200 per month | Yes | GPT-6.2 Sol only |
| Pro $500 | $500 per month | Yes | GPT-6.2 Sol and Astra, plus GPT-6 Astra and GPT-6.1 Sol |
| Enterprise / Edu | Contact sales | Yes | Both GPT-6.2 models, if the workspace is eligible and the owner turns it on |
Not sure which plan you are on, or which one fits you? See our ChatGPT plans comparison.
Tip: Some OpenAI pricing pages still say “Ultrafast mode access on Pro $500” only. The 17 November changelog and the GPT-6.2 release notes both confirm Pro $200 now gets Ultrafast with Sol.
How Much Usage Ultrafast Costs
ChatGPT Work and Codex share one usage pool. Ultrafast drains that pool faster than Standard, by a fixed multiplier.
| Speed and model | Included plan usage | Credits / pay as you go |
|---|---|---|
| Fast, GPT-6.2 Astra or Sol | 2x | 2x |
| Fast, GPT-6.1 Sol or Terra, GPT-6 models | 2.5x | 2x |
| Ultrafast, GPT-6.2 Astra or Sol | 6x | 3x until 28 Feb 2027, then 6x from 1 Mar 2027 |
| Ultrafast, GPT-6 Astra | 8x | 6x |
| Ultrafast, GPT-6.1 Sol | 8x | 6x |
What 6x means in practice
One Ultrafast task on GPT-6.2 Sol uses about as much of your allowance as six Standard tasks. So a five-hour window that fits 60 Standard Sol tasks would fit about 10 in Ultrafast.
You get up to 4.5x speed for 6x the usage. That is a fair trade only when your time is worth more than the extra allowance. On Pro $500, Ultrafast uses your included allowance first, then any credits you have.
Warning: The cheaper 3x credit rate is a launch offer. From 1 March 2027 (Pacific time), credits and pay-as-you-go Ultrafast on GPT-6.2 go to 6x. Budget for that now if your team relies on it.
For a full breakdown of five-hour windows and weekly caps, see our ChatGPT Codex usage limits guide.
How to Turn On Ultrafast in ChatGPT Work and the Codex App
The control is a lightning bolt in the composer. It works the same way in ChatGPT Work on the web and desktop and in the Codex app.
- Pick a supported model. Choose GPT-6.2 Sol (Pro $200 or higher) or GPT-6.2 Astra (Pro $500, Enterprise or Edu).
- Open the model or effort picker. It sits at the bottom of the composer.
- Select the lightning bolt. If Ultrafast is available to you, this opens a speed menu.
- Choose Ultrafast. The menu lists Standard, Fast and Ultrafast. Pick Ultrafast and send your prompt.
- Or use the advanced picker. Open it and select Speed to set the same option.
- Switch back when done. Return to Standard for long, low-urgency work so you save allowance.
If you only see Standard and Fast, the lightning bolt flips between those two at once. No menu opens. That usually means your plan or model does not support Ultrafast.
On mobile
In ChatGPT Work on iOS or Android, open the model picker and use the speed setting there. The options match the desktop ones.
How to Turn On Ultrafast in the Codex CLI
The Codex CLI uses slash commands for speed. You must be signed in with your ChatGPT account for plan usage to apply. New to the CLI? Start with our Codex guide.
- Update the CLI. Run
npm i -g @openai/codex@latestso the GPT-6.2 models and speed commands appear. - Pick a model. Type
/modeland choose GPT-6.2 Sol or GPT-6.2 Astra. - See what your model supports. Type
/to list the commands available for the current model. - Turn on Ultrafast. Run
/ultrafast. The CLI remembers this choice for later sessions. - Turn it off. Run
/ultrafastagain to go back to Standard. Use/fastfor the middle tier. - Show it in the footer. Run
/statuslineto add the speed mode to your status line.
Since Codex CLI 0.164.0 (23 October 2026), you can also set speed inline with the /model command arguments, together with the model and reasoning effort.
Config file switches
The Codex config reference lists two feature flags. Both are on by default, so most people never touch them.
# ~/.codex/config.toml
[features]
fast_mode = true # show Fast controls when the model supports them
ultrafast_mode = true # show Ultrafast controls, separate from fast_mode
Set ultrafast_mode = false if you want to hide the option so nobody on a shared machine burns allowance by mistake.
Note: If you sign in to Codex with an API key instead of ChatGPT, you pay API token rates. The plan multipliers above only apply with ChatGPT sign-in. OpenAI’s docs mention Fast for the IDE extension, but do not list Ultrafast steps there yet.
Enterprise and Edu: Enabling Ultrafast for a Workspace
In business workspaces, Ultrafast is off by default. A workspace owner has to opt in. These are the requirements:
- Enterprise contract: needs a credit-based or USD usage-based agreement.
- Edu contract: needs a credit-based plan.
- Models: owners must first enable the GPT-6.2 models members will use.
- Permission: Ultrafast is a separate permission in workspace settings. Owners can grant it to chosen users or the whole workspace.
- Spend controls: per-user spend limits still apply to Ultrafast usage.
There is also a data residency limit. Ultrafast on GPT-6.2 Astra, GPT-6.2 Sol and GPT-6 Astra runs in the US only. GPT-6.1 Sol Ultrafast also supports Europe (EEA and Switzerland), so EU-bound teams may need to stay on that model.
Ultrafast in the OpenAI API
Yes, there is an API version. It is a service tier on the Responses API. You set service_tier to "ultrafast" in the request body.
from openai import OpenAI
client = OpenAI()
response = client.responses.create(
model="gpt-6.2-sol",
service_tier="ultrafast",
input="Write a Python function that parses ISO 8601 dates."
)
print(response.output_text)
Supported API models are gpt-6.2-astra, gpt-6.2-sol, gpt-6-astra and gpt-6.1-sol. Ultrafast has its own rate limits, separate from Standard and Fast.
API prices per 1M tokens
| Model and tier | Input | Cached input | Output |
|---|---|---|---|
| GPT-6.2 Sol, Standard | $2.00 | $0.10 | $10.00 |
| GPT-6.2 Sol, Fast | $4.00 | $0.20 | $20.00 |
| GPT-6.2 Sol, Ultrafast (until 28 Feb 2027) | $6.00 | $0.30 | $30.00 |
| GPT-6.2 Sol, Ultrafast (from 1 Mar 2027) | $12.00 | $0.60 | $60.00 |
| GPT-6.2 Astra, Standard | $8.00 | $0.40 | $40.00 |
| GPT-6.2 Astra, Ultrafast (until 28 Feb 2027) | $24.00 | $1.20 | $120.00 |
| GPT-6.2 Astra, Ultrafast (from 1 Mar 2027) | $48.00 | $2.40 | $240.00 |
A fun detail: Sol in Ultrafast ($30 output, launch rate) is still cheaper per output token than Astra at Standard speed ($40). For more on tiers, see our OpenAI API pricing guide.
Use WebSockets for agents
OpenAI strongly recommends a persistent WebSocket connection for Ultrafast, especially for agents that make many quick tool calls. Over plain HTTP, connection overhead can eat much of the speed gain.
In WebSocket mode, include service_tier in each response.create event. API Ultrafast data residency for the GPT-6.2 models is also limited to the US.
When Ultrafast Is Worth It (and When It Is Not)
Since Ultrafast costs 6x usage for up to 4.5x speed, you get less work per unit of allowance. Use it where waiting is the real cost.
Good uses
- Live pair programming: quick edit-test loops where you watch every step.
- Debugging under pressure: a production bug where minutes matter.
- Demos and meetings: when a room full of people waits on the answer.
- Voice or chat products (API): user-facing apps where latency hurts the experience.
- Short, high-value prompts: small tasks where 6x of a tiny amount is still tiny.
Skip it for
- Long background agents: runs you check later gain nothing from speed.
- Big refactors: they eat allowance fast and you hit limits early.
- Tool-heavy tasks: time spent running tests or builds does not speed up.
- Batch API work: the Batch tier costs half of Standard, the opposite trade.
If you also use Anthropic tools, the idea is similar to Claude’s speed option. Our Claude fast mode guide covers how that one is priced.
Troubleshooting: Ultrafast Not Showing or Not Working
- No lightning bolt menu, only a toggle: your current model or plan supports Fast but not Ultrafast. On Pro $200, switch the model to GPT-6.2 Sol.
- Ultrafast missing for GPT-6.2 Astra on Pro $200: this is expected. Astra Ultrafast needs Pro $500, Enterprise or Edu.
- Option missing on Enterprise or Edu: ask your workspace owner. Ultrafast is off by default and needs a credit-based or usage-based contract.
/ultrafastnot found in the CLI: update withnpm i -g @openai/codex@latest, pick a GPT-6.2 model, and checkfeatures.ultrafast_modeis not set to false.- Charged API rates in Codex: you are signed in with an API key. Log out and sign in with ChatGPT to use plan usage.
- Hitting limits much sooner: that is the 6x multiplier. Switch back to Standard or Fast for long tasks.
- Not much faster in practice: tool calls, tests and network round trips do not speed up. In the API, move to WebSockets.
- Blocked for EU data residency: GPT-6.2 Ultrafast is US only. Use GPT-6.1 Sol Ultrafast, which supports the EEA and Switzerland.
If ChatGPT itself is down or failing, see our ChatGPT not working fix guide.
Frequently Asked Questions
Does Ultrafast make answers worse?
No. Ultrafast runs the same model at the same reasoning effort. Only the token speed and the usage cost change.
Can ChatGPT Plus users get Ultrafast?
No. Plus users can use Fast mode, but Ultrafast starts at Pro $200 (Sol only) and Pro $500 (Sol and Astra), plus eligible Enterprise and Edu workspaces.
How much faster is Ultrafast than Standard?
OpenAI says up to 4.5x Standard speed on GPT-6.2 Astra and Sol. Older GPT-6 Astra in Codex reaches up to 8x.
How much more usage does Ultrafast use?
On GPT-6.2 models it uses 6x your included plan usage. Credits and pay-as-you-go cost 3x until 28 February 2027, then 6x from 1 March 2027.
What is the Codex CLI command for Ultrafast?
Run /ultrafast to turn it on and run it again to go back to Standard. Use /fast for the middle tier.
Is Ultrafast the same as the old Priority processing?
No. Priority processing was renamed Fast mode on 30 July 2026. Ultrafast is a separate, faster and pricier tier above it.
Can I use Ultrafast through the API without a Pro plan?
Yes. In the API you pay per token, so set service_tier to "ultrafast" on a supported model. Check your organization’s Ultrafast rate limits first.
Sources
- ChatGPT and Codex changelog (17 November 2026 entry)
- OpenAI: What’s new in GPT-6.2 Astra and Sol
- OpenAI: Codex speed options (Fast and Ultrafast)
- OpenAI: Codex pricing and Fast/Ultrafast multipliers
- OpenAI API: Ultrafast mode guide
- OpenAI API pricing by service tier
- Codex configuration reference
- The Next Web: GPT-6.1 Sol and Ultrafast at DevDay 2026
- The Batch: Sol Nearly Eclipses Astra




