HomeFull speed ahead. Keep track of advancements in every frontier

TheTechno Optimist.

CodexOpenAI—StargateOpenAI—OpenAI—GPT modelsOpenAI—OpenAI—gpt-ossOpenAI—CodexOpenAI—GPT modelsOpenAI—
Codex
Courtesy of OpenAI · Press kit, editorial use (opens openai.com)

Codex

by OpenAIUS

ScalingStage 5 of 5

OpenAI's coding agent, in the terminal, code editors and the cloud, running GPT-5.x-Codex models.

Updated 5 Feb 2026Checked 25 Sep0 updates this week

Milestones

No announced next step
  1. Original Codex model powers GitHub CopilotAug 2021Complete.
  2. Codex CLI open-sourced16 Apr 2025Complete.
  3. Codex cloud agent in ChatGPT (research preview)16 May 2025Complete.
  4. GPT-5-Codex model15 Sep 2025Complete.
  5. GPT-5.3-Codex modelFeb 2026Complete.

Most important updates

  • 5 Feb 2026
  • 2 Feb 2026
  • 15 Sep 2025
  • 16 May 2025
  • 10 Aug 2021

Current obstacles

  • Checking agent-written codeAgents can pass tests while adding subtle bugs or security holes. Humans must review unless checks are automated.

Physics limits

  • Errors compound over long tasksIf each step succeeds 99% of the time, a 100-step task succeeds only about 37% of the time (0.99^100). Agents need near-perfect per-step reliability to work for hours.
  • No program can prove code correct in generalRice's theorem: no algorithm can decide non-trivial properties of every program. Agents lean on tests, and code that passes weak tests can still be wrong.
  • Large codebases overflow the contextBig repositories hold tens of millions of tokens, far beyond a 1M-token window, so the agent must search and summarise and can miss distant dependencies.

How it works

4 parts
Server aisle in a Microsoft data centre: each Codex cloud task runs in its own sandbox on servers like these
Server aisle in a Microsoft data centre: each Codex cloud task runs in its own sandbox on servers like thesePhoto: Robert Scoble · CC BY 2.0 (opens commons.wikimedia.org)
Sandbox

A private copy of your code

Each cloud task runs in its own isolated container loaded with your repository, with internet off by default, so the agent can't touch anything else.

Loop

Read, edit, run, repeat

The model reads files, writes changes and runs tests and linters in a loop, fixing its own mistakes until checks pass, then proposes a pull request.

Model

Coding-tuned GPT models

GPT-5.x-Codex models get extra reinforcement learning on real coding tasks, and adjust how long they think, from seconds to hours.

Guidance

Instructions in the repo

An AGENTS.md file tells it how the project is built and tested. It shows terminal logs and test output so a person can check its work.

Update log

6 updates

Thu 5 Feb

  • Major: BlogSoftware

Mon 2 Feb

  • Minor: BlogSoftware

Mon 15 Sep 2025

  • Major: BlogSoftware

Fri 16 May 2025

  • Major: BlogSoftware

Wed 16 Apr 2025

  • Minor: BlogSoftware

Tue 10 Aug 2021

  • Major: BlogSoftware

About OpenAI

The team behind Codex

OpenAI

ChatGPT, GPT models, Codex, Stargate

OpenAI makes ChatGPT and the GPT models behind it. ChatGPT's launch in November 2022 made AI chatbots mainstream. It also builds huge data centers through the Stargate project.

  • Makes ChatGPT and the GPT-5.x models. GPT-5.6 came out in July 2026.
  • Raised $122B at an $852B valuation in March 2026. Amazon, NVIDIA and SoftBank were the biggest investors.
  • Lining up tens of gigawatts of computing power through Stargate and chip deals with NVIDIA, AMD and others.

openai.com (opens openai.com)Company profile →

Founded
201511 yrs
Headquarters
United States
Status
Private
Valuation
$852BprivateMar 2026
Raised
$179B4 rounds
Last round
Venture round · $122BMar 2026$852B post
Works in
AIFrontier modelsComputing
Coverage
4 programs · 24 updateslatest 9 Jul 2026checked 25 Sep
Lead investors
AmazonNVIDIASoftBankThrive CapitalMicrosoft
Partners
Broadcom
Investors
MicrosoftSoftBankAmazon
Suppliers
OracleNVIDIAAMDCoreWeaveCerebras
People
Sam AltmanCEO and co-founderGreg BrockmanPresident and co-founderJakub PachockiChief Scientist
Primer