OpenAI introduced the Ultrafast service tier during the 2026 Developer Day. This new offering increases token generation speeds by up to 8 times in Codex and up to 6 times for API calls. The service is designed to provide real-time responses while maintaining the capabilities of the GPT-5.6 Sol model, targeting use cases like live customer support, market alerts, and interactive coding.
The GPT-6 Astra Ultrafast model achieves a generation speed of 300 tokens per second. It is currently available via API, and access is provided to Pro 500 subscribers and enterprise users through ChatGPT Work and Codex. OpenAI also announced plans to release the GPT-6.1 Sol Ultrafast model in the future.
Pricing for the service varies based on context length. For short context under 272K tokens, input costs $60.00 per 1M tokens, while output costs $300.00 per 1M tokens. For long context exceeding 272K tokens, these rates increase to $120.00 and $450.00 per 1M tokens respectively. Cached input costs are significantly lower, starting at $6.00 per 1M tokens for short context.
