Codex isn’t only a chat box with a repo attached. At DevDay 2026 (29 September 2026), OpenAI stacked model speed, agent APIs, and a security scanning cloud under the same developer story. This is the “what’s underneath” post — denser on purpose, still in plain English.
I’m skipping absolute dollar prices OpenAI didn’t put on the recap. Relative claims stay labeled as OpenAI’s numbers.
GPT‑6.1 Sol — near-Astra pitch at one-fifth the tokens
OpenAI calls GPT‑6.1 Sol a major upgrade to GPT‑6 Sol, strong at agentic coding, computer use, and professional work, with “near-Astra intelligence” at one-fifth of Astra’s standard input and output token prices.
Access: API, Plus, Pro, Business, Enterprise, and Edu — per the recap. Launch note: Introducing GPT‑6.1 Sol
What I’d do with that: if Astra felt too spendy for everyday coding agents, Sol is the plan to A/B on the same task. I’d measure task success + cleanup time, not only tokens. “Near-Astra” is their phrase; your PR quality is the score that matters.
Ultrafast — premium speed, gated access
Ultrafast is a paid speed tier. OpenAI’s recap claims up to 8× faster token generation (300 tokens/sec) in Codex and up to 6× in the API. Treat those as vendor performance marketing until you see it on your workload.
Availability called out:
- GPT‑6 Astra Ultrafast: available now in the API, and in ChatGPT Work and Codex on Pro 500 and Enterprise
- GPT‑6.1 Sol Ultrafast: coming soon
Docs: Ultrafast mode
Pro 500 itself is described as the highest usage allowance — 25× ChatGPT Plus — and includes Ultrafast access. Absolute monthly price wasn’t on the DevDay recap; confirm on OpenAI’s Pro tiers help article before we print a rupee or dollar figure.
Agents API (with computer use) and Decisions API (Luna)
Agents API: recap says it now supports computer use — agents operating software through the UI — and brings Codex-style multi-agent abilities, tool search, tool calling, and context compaction into apps, with OpenAI running the infrastructure. Access for computer use is listed through the API and in Codex / ChatGPT Work on Pro 500 and Enterprise. Overview: Introducing the Agents API
Community write-ups call Agents API a public beta with hosted execution and memory. I’d keep “public beta” only if the Agents launch post still says it on publish day — re-check that URL before we lock the phrase.
Decisions API (Luna): real-time choices by focusing Luna on a fixed set of user-defined questions with finite predefined answers (text/image context). Aimed at classify / route / pick the next agent step. Recap: limited preview today, broader release “in the coming days.” That second part can slip — date-check before homepage placement.
There’s also Amazon Bedrock Managed Agents powered by OpenAI for agents that run in AWS. Commercial detail lives on AWS’s side: Bedrock Managed Agents for OpenAI.
Codex Security Cloud
Security Cloud sits closer to the Codex story than a generic “AI security” blog:
- Scan entire GitHub repositories on demand or on a schedule
- Ongoing checks on new commits
- Investigate findings, dedupe, prepare fixes in the cloud (works with the laptop closed)
- Includes access to models offered through Daybreak Blue without a separate Daybreak application
Access: all Pro, Business, Enterprise, and Edu users on desktop and web (plugin), per the recap. Docs: Security setup · Daybreak
No scan quotas or false-positive rates on the recap. I’d enable it on a non-critical repo first and read every proposed fix like a junior engineer’s PR.
How I’d sequence adoption
- Sol on one coding agent workflow — compare quality vs cost against whatever you use now.
- Ultrafast only if you’re already on Pro 500 / Enterprise (or API Ultrafast) and latency is the actual bottleneck.
- Security Cloud on a side repo before the monorepo.
- Agents / Decisions when you have a clear UI-automation or routing problem — not as a toy.
Bottom line
DevDay’s Codex story isn’t one button. It’s cloud environments, everyday CLI/review tools, cheaper-faster model options, and a security scanner that can work while you’re offline. I’d pick one layer this week and write down what broke. That’s how this stuff becomes useful instead of slideware.
