Breaking

OpenAI Adds Ultrafast Speed Tier for Codex and GPT-6 Astra API

September 29, 2026 at 13:59 EDT

Software Dev & Coding
AI Agents
Infra & Chips

OpenAI has introduced Ultrafast as a premium speed tier for Codex and the API, turning faster token generation into an explicit product and billing choice. The company framed Codex around up to 8x faster generation, reaching 300 tokens per second, while current API pricing lists GPT-6 Astra Ultrafast at six times Standard pay-as-you-go rates.

Sep. 29, 2026 | OpenAI

Ultrafast moves into Codex

OpenAI is making faster GPT-6 Astra inference a premium tier, led by up to 8x faster token generation in Codex and a higher-cost API path for speed-sensitive agents.

faster token generation in Codex

300

tokens per second peak claim

paid credits and pay-as-you-go rate

$500

Pro plan level with Astra Ultrafast

The speed jump, drawn to scale

One Standard unit beside eight Ultrafast units shows the Codex token-generation comparison.

Standard

Ultrafast

Takeaway: faster tokens can make interactive coding feel immediate, but full agent time still depends on tools, tests, approvals, and network overhead.

How teams turn it on

Choose GPT-6 Astra

Set service_tier to ultrafast

Keep agent loops on WebSockets

The tier is meant for work where latency is worth the premium, not every batch job.

Why developers want it

Less waiting between a prompt, a generated patch, and the next edit can make coding agents feel more conversational and easier to keep in flow.

What limits the gain

The same speed can consume limits faster, and jobs blocked on tests, tools, approvals, or network trips may not finish 8x sooner.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.