OpenAI's latest update cycle is not one single product launch. It is a pattern: faster high-end inference for developers, more ways for ChatGPT and Codex users to manage capacity, and more segmentation between casual, power, and business usage.
The headline feature is Ultrafast, a new OpenAI API service tier that runs GPT-5.6 Sol up to 14x faster than Standard processing. OpenAI says the preview is powered by Cerebras and can generate up to 750 output tokens per second.
That speed is not arriving as a blanket ChatGPT toggle. OpenAI says Ultrafast is launching first in the API, in limited preview for a select group of customers, with broader access planned as capacity grows.
What changed with Ultrafast
OpenAI frames Ultrafast as a way to make frontier intelligence practical in workflows where delay changes the product itself.
The company names incident response, financial research, customer support and voice, commerce, and live research as early use cases. The common thread is not just "faster answers." It is interaction: the model can read evidence, answer, let a user adjust direction, then respond again without turning a task into an overnight batch job.
For developers, the important distinction is that Ultrafast is an API service tier, not the same thing as Codex Fast mode.
Codex already has Fast mode for supported models in ChatGPT-connected Codex environments. OpenAI's Codex speed docs say Fast mode increases supported model speed by 1.5x and consumes credits at a higher rate. For GPT-5.6 and GPT-5.5, that multiplier is listed as 2.5x the Standard credit rate.
Ultrafast is more ambitious: GPT-5.6 Sol in a limited API preview, running at much higher throughput, with Cerebras supplying the low-latency inference layer.
Usage is becoming a credit product
The other practical update is about limits.
OpenAI's current Codex pricing page says ChatGPT Plus and Pro users who hit their usage limit can purchase additional credits to keep working without upgrading their existing plan. Business, Edu, and Enterprise customers on flexible pricing can purchase additional workspace credits.
That matters because the public plan table still separates short-window usage from longer pressure on the account. For Plus, the page lists local Codex messages in a five-hour window and says additional weekly limits may apply. It also says users approaching limits can switch to GPT-5.6 Luna to stretch remaining allowance.
The Codex rate card makes the direction clearer. OpenAI says Codex pricing has shifted from per-message estimates to token-based credit rates across current Plus, Pro, Business, Enterprise, Edu, Health, Gov, and teacher plans, with usage based on input tokens, cached input tokens, and output tokens. It also says Codex, ChatGPT Work, ChatGPT for Excel, and Workspace Agents draw from the same agentic usage and credit pool when available on a plan.
So the new model is less like "you get a fixed number of chats" and more like a managed work budget: higher-end models, larger context, more output, more agents, and faster modes can consume more of the pool.
ChatGPT Business gets a premium seat
OpenAI is also preparing Premium seats for ChatGPT Business.
The new seat is aimed at the most active people inside a business workspace. OpenAI says Premium will provide 5x more usage than Standard, remove the five-hour usage limit, and reset usage weekly. It is priced at $125 per user per month, or $100 per user per month annually.
Standard Business seats remain at $25 monthly, or $20 monthly on annual billing. Workspaces can mix Standard and Premium seats, which means a team can give heavier users more room without upgrading everyone.
There is a launch promotion too. OpenAI says the first 10,000 eligible Business customers can receive $100 in workspace credits, equal to 2,500 credits, for each Premium seat they add, up to five seats. The promotion ends on August 20, 2026.
Other OpenAI updates this week
The August 14 ChatGPT release notes added several smaller but useful changes.
Interactive quizzes are now available across consumer ChatGPT plans and Edu plans on web and mobile. Eligible unshared Projects can switch between default memory and project-only memory after creation. Paid users may also see homepage suggestions based on conversation history and connected tools.
OpenAI also expanded platform support. The ChatGPT desktop app for Linux is now in public preview globally on Ubuntu 24.04 LTS, Ubuntu 26.04 LTS, Debian 13, Fedora 43, and Fedora 44. The Linux app supports ChatGPT and Codex, plus browser actions inside the built-in browser or Chrome. Controlling other desktop apps is not yet supported.
On August 13, OpenAI added Google Drive to Library for eligible Plus, Pro, Enterprise, Edu, Healthcare, and Business users on web. Users can browse Drive files and folders from Library, pull files into chat with the composer or @mentions, and keep Google Docs, Sheets, and Slides open beside the conversation where supported.
OpenAI's ads pilot also widened on August 11. ChatGPT Ads launched in the United Kingdom, Mexico, Brazil, Japan, and South Korea after earlier U.S. and other-market testing. OpenAI says ads remain limited to logged-in adult users on Free and Go tiers, and do not appear for Plus, Pro, Business, Enterprise, or Education tiers.
For enterprise security teams, OpenAI also announced that Daybreak Red and Daybreak Blue are available through Amazon Bedrock for eligible customers enrolled in Daybreak Access. That brings OpenAI's frontier cybersecurity models into AWS procurement, governance, and operational workflows.
Why it matters
OpenAI is separating two problems that used to feel blended together: how fast a strong model can respond, and how much sustained work a user or team can run.
Ultrafast answers the latency side. It is built for products where a frontier model has to keep pace with real-time work: calls, incidents, shopping decisions, financial analysis, and interactive research loops.
Credits and Premium seats answer the capacity side. Heavy users no longer fit neatly into a single flat subscription bucket, especially when Codex and ChatGPT Work can run long tasks across files, tools, codebases, browsers, and connected apps.
The result is a more enterprise-shaped ChatGPT stack. Casual users get simpler access and, in some regions, ads on lower tiers. Power users get visible usage controls and credit options. Teams get seat tiers and shared workspace credits. Developers get faster API service classes.
Our take
This is OpenAI turning speed and usage into separate product levers.
For builders, Ultrafast is the most strategically interesting update because it makes GPT-5.6 Sol plausible in user-facing products that could not tolerate slow frontier-model latency. For ChatGPT and Codex users, the more immediate change is practical: usage limits are becoming something you can monitor, stretch with smaller models, or top up with credits depending on your plan.
The tradeoff is complexity. OpenAI now has model tiers, service tiers, speed modes, credits, short-window limits, weekly limits, seat types, and workspace controls. That is more flexible, but it also means serious users need to watch the usage dashboard as closely as they watch model choice.
If you are using ChatGPT Work or Codex for real projects, the safest workflow is to treat GPT-5.6 Sol as the expensive high-power setting, use Luna for routine iterations, and reserve fast or extra-credit spending for the tasks where time actually changes the outcome.