OpenAI makes GPT-6 Astra and GPT-6.1 Sol about 50 percent faster

What's new

OpenAI has adjusted how fast its two models, GPT-6 Astra and GPT-6.1 Sol, respond by default. According to an announcement from October 5, 2026, from Thibault Sottiaux, who leads Codex development at OpenAI, speeds go from roughly 30 to roughly 50 tokens per second. A token is a small chunk of text, so the text appears on screen noticeably faster. Independent measurements cited in the coverage land at 50.4 tokens per second for GPT-6.1 Sol and 47.7 for GPT-6 Astra.

The speedup applies to subscribers, both in OpenAI's own products and in partner programs where you sign in with your ChatGPT account. OpenCode, Pi, Amp and Devin are mentioned.

What's smart about it

It's a free improvement. You don't have to switch models, turn anything on or pay extra. It's most noticeable when the AI writes long answers or works as a coding assistant, where you would otherwise be sitting and waiting for it to finish.

OpenAI also offers two special options for those who want even more speed. Fast mode can be enabled in the API (the technical access for developers) by adding a single setting, service_tier=fast. Ultrafast is a more expensive option that, according to the coverage, can reach up to 300 tokens per second. Ultrafast was shown off in August and is so far only available to invited customers.

Cheaper or better?

Prices are unchanged. In the API, GPT-6.1 Sol costs $2 per million tokens for input and $10 per million tokens for output. GPT-6 Astra, the big flagship, costs $10 per million tokens in and $50 per million tokens out. That makes Sol five times cheaper than Astra. What the subscriptions cost per month, and whether the improvement reaches Denmark and the EU at the same time, is not stated in the material I have been able to find.

Comparisons with competitors such as Google and Anthropic are also not provided in the sources, so I can't say whether OpenAI is now the fastest. It's worth knowing that we are relying on a single detailed source here and not on OpenAI's own page, so the figures should be taken with a small grain of salt.

Who it's good for

The speed is especially good news for developers and companies that use AI for coding and long tasks, because waiting time drops. It also benefits everyday users who use AI to write drafts, summaries and reports, where you want to see the result quickly.

SD Times' news roundup from October 9 also mentions a new MCP server from Functionalize and lower prices for smaller companies on Oracle's Java SE Universal Subscription. I have not been able to confirm details about those, so they are not included here.