OpenAI unveils a $500 Pro 500 subscription and Ultrafast mode

At the DevDay 2026 conference in San Francisco, OpenAI introduced the Pro 500 subscription at $500 a month and a premium Ultrafast mode. In Codex, token generation speeds up by up to eight times, to roughly 300 tokens per second. In the API the speedup reaches six times, but the price there is also six times the standard rate.

For GPT-6 Astra, the accelerated mode in the API costs $60 per million input tokens and $300 per million output tokens. Astra also has an intermediate Fast mode: it is twice as expensive as the standard one, while running 2.5 times faster. Ultrafast is already available in the API, and also in ChatGPT Work and Codex for enterprise customers and Pro 500 subscribers.

OpenAI calls Pro 500 the tier with the largest usage limit among its subscriptions, but it does not disclose specific volumes, so comparing it with the previous Pro comes down mostly to price.

The old Pro's limits were cut back

The limits of the previous $200 Pro were cut back at the same time. In ChatGPT Work and Codex it now gives ten Plus volumes instead of twenty, and GPT-6 Pro messages in chat were halved, from 200 to 100 per week.

Existing subscribers will keep their previous limits during a transition period and will receive a one-time credit.

What else was announced at DevDay

The speedup is only part of the DevDay 2026 announcements. Codex received reusable cloud environments, voice control in the CLI, and code review viewing in the desktop app. For security teams, Codex Security Cloud with repository scanning was launched. Computer Use and the Decisions API on the GPT-6 Luna model were added to the Agents API, and the Bedrock Managed Agents platform on AWS is adopting key Agents API capabilities.

Translation editor: Салтанов А.