OpenAI DevDay: faster Codex and GPT-6.1 Sol

On Sept. 29, 2026, the company’s recap describes faster code generation, a model cheaper than Astra, and tools for review and parallel work.

On Sept. 29, 2026, in San Francisco, OpenAI published the official recap of OpenAI DevDay 2026. The same day, @OpenAIDevs posted a thread titled “Code faster, review changes, and keep parallel work moving,” with a roundup for people who write software and a link to the recap. The company says there were “more than 20 major announcements across ChatGPT, Codex, our models, and entirely new forms of working with AI.” CNBC, reporting live from the venue, also recorded “more than 20” products and features in the keynote, which Sam Altman closed.

The recap matters if you pay for Codex or the API. It is OpenAI’s own list of what got faster, cheaper, or new in the coding tools. This story covers that package.

First: what Codex is, and what a token is

Codex is OpenAI’s line of tools for writing and reviewing code with the model, on a computer, a phone, or a terminal. A token is a chunk of text the model reads or writes. The API, the programming interface that connects the model to an app, bills per million tokens. When the company says “faster” or “one-fifth the price,” it is usually talking about that unit, not the wall-clock time of a whole project.

Speed at OpenAI DevDay, according to OpenAI

The recap describes Ultrafast as a premium speed tier. OpenAI says token generation is up to 8 times faster in Codex, at 300 tokens per second, and up to 6 times faster in the API. The baseline is standard generation, measured by the company, on Sept. 29, 2026. That is not end-to-end task time, and it is not an independent measurement.

In that recap, GPT-6 Astra Ultrafast was already in the API and in ChatGPT Work and Codex on the $500-per-month Pro plan and Enterprise. Workspaces with inference residency outside the United States, meaning they require the model to run in another country, are not eligible. GPT-6.1 Sol Ultrafast was still “coming soon.” In API docs read on Oct. 8, 2026, Ultrafast (service_tier: ultrafast) is listed as the fastest tier and “broadly available” for GPT-6 Astra and GPT-6.1 Sol.

GPT-6.1 Sol and the one-fifth price claim

GPT-6.1 Sol arrived a week after GPT-6 Sol, which launched on Sept. 22, 2026. OpenAI presents it as an upgrade that “nearly matches GPT-6 Astra’s intelligence on agentic coding, computer use, and professional work at one-fifth of Astra’s standard input and output token prices.” TechCrunch’s Aisha Malik reported the claim and noted there was no GPT-6.1 Astra. The “nearly Astra” intelligence claim is the vendor’s own benchmark.

On OpenAI’s table from Sept. 29, 2026, Astra costs $10 input, $50 output, and $1 cached input per million tokens. GPT-6.1 Sol, API id gpt-6.1-sol, costs $2, $10, and $0.10. Two divided by ten is one-fifth; ten divided by fifty is the same. The $0.10 cache is 95% below Sol’s own standard input of $2, not one-fifth of Astra’s cache. The one-fifth figure is standard input and output versus Astra, not cache and not Ultrafast.

At announcement, the model was in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu, and in the API. It was not yet in Chat.

CLI, cloud, review, and plugins

The Codex CLI, the terminal program, gained voice to start tasks and an /agents view to watch several jobs at once. OpenAI says it is on all plans. Codex in the cloud describes reusable environments on a computer, a phone, or remotely, for Plus, Pro, Business, Healthcare, Education, and Enterprise.

In the ChatGPT desktop app, Code Review summarizes diffs, the differences between file versions, and lets you ask Codex about GitHub pull requests (generally available, according to the docs) or GitLab merge requests (preview), with automatic reviews in the cloud. Plugin extensions open interactive panels, file viewers, and a “home” in the sidebar. The recap says all plans; the docs say web support is still “coming soon” for Free and Go.

Limits and what we still don’t know

We did not test CLI voice, automatic review, or cloud environments. Speed and “nearly Astra” claims come from OpenAI, which sells the product. CNBC and TechCrunch covered the event without auditing those figures.

We still don’t know whether 300 tokens per second shows up in real use, whether Sol holds up to Astra outside the company’s benches, or when the model reaches Chat. Ultrafast pricing was not in the table this check read.

If you already pay for Codex or the API, you can check the recap and the docs. The other announcements from that day stay out.

Sources

Official

Press

First: what Codex is, and what a token is

Speed at OpenAI DevDay, according to OpenAI

GPT-6.1 Sol and the one-fifth price claim

CLI, cloud, review, and plugins

Limits and what we still don’t know

Sources

Sources

Markers