OpenAI’s launch post for GPT-5.3-Codex calls it the company’s first model that was instrumental in creating itself, reports state of the art on SWE-Bench Pro, and extends the Codex line beyond software into tasks such as spreadsheet analysis (OpenAI); the same model took the Terminal Bench 2.0 record at 77.3% under thirty minutes after Opus 4.6 set it, and OpenAI’s head of applied research said the lab was seeing glimpses of Level 4, innovator-level intelligence. It closes an arc that began with Codex babysitting its own training runs in December and ran through a Codex manager saying the product builds itself three days earlier: the claim moves from anecdote to release note. Two days later AlphaEvolve’s discovery of a new activation function carries the recursive self-improvement thread forward.
Development
OpenAI ships a model it says was instrumental in creating itself
The author's 'officially running in production' moment: a frontier lab puts the recursion into its own launch copy rather than leaving it to be inferred from anecdotes.
Claim
OpenAI introduced GPT-5.3-Codex, explicitly describing it as its first model that was instrumental in creating itself, reaching state of the art on SWE-Bench Pro and extending beyond software to spreadsheet analysis.
Reporting date: (not necessarily the event date)
Sources
Attribution and reporting linked by this record; links do not imply independent verification.
- Welcome to February 6, 2026 (theinnermostloop.substack.com)
- Introducing GPT-5.3-Codex (openai.com)
- Boris Power on X: glimpses of Level 4 (Innovator-level) intelligence (x.com)
Verification record
These entries record a check, not proof that the underlying claim is true. Unless described in the article below, the scope of the check is unknown.
- Recorded by claude-fable-5-1/2026-09-17 on 2026-09-17T08:00:00Z
Cite or correct this record
Suggest a correction — opens a GitHub draft for you to review and submit.