The autonomous software engineering agent that writes, tests, and ships code — in parallel
83
Overall score
45
Heat score
Inputs
Natural Language Task Description, GitHub Repository, Codebase Context, AGENTS.md Instructions, Bug Report, Feature Spec, Pull Request, Screenshot, CI Pipeline Log
Outputs
Written Code, Edited Files, Pull Request, Bug Fix, Test Suite, Codebase Explanation, Documentation, Refactored Code, Pipeline Autofix, Implementation Plan
AI Type
Agentic AI
Model Architecture
GPT-style Decoder
Daily Prompts
N/A
Context Length
N/A
Accuracy
95%
Content
93%
Reasoning
96%
Company
OpenAI
Founded
2015
HQ
San Francisco, California, USA
Employees
3.5K
Total Raised / Total Funding
$11.3B
Revenue
$3.7B
Valuation
$157B
ARR
N/A
CEO
Sam Altman
Estimated Paid Users
1M
Current estimate
Total Earnings Till Date
N/A
No monthly delta yet
Market Share
8.4%
Current share
Average Session
90
Per active user
Hallucination Rate
5%
Model quality signal
Growth Rate
+21.43%
Monthly active users
Burn Rate
N/A
Total expenses / years active
Paid User Gain
+33.33%
Monthly paid user trend
$0
Total Loss
$0
Total Profit
$0
Accuracy
95%
Context
93%
Reasoning
96%
Safety
90%
HumanEval
28.8%
SWE-bench Verified
72.1%
SWE-bench Verified
74.9%
Type: Code
Description: Original Codex model. GPT-3 descendant trained on 159 GB of Python code from 54 million public GitHub repositories. Supported 12+ programming languages. Powered GitHub Copilot's first technical preview. Achieved HumanEval pass@1 of 28.8% (12B parameter version). Deprecated March 2023.
Architecture: GPT-style Decoder
Type: Reasoning
Description: First model of the 2025 Codex agent era. A version of OpenAI o3 fine-tuned for software engineering via reinforcement learning on real-world coding tasks across diverse environments. Generates code that mirrors human style and PR conventions. Scored 72.1% on SWE-bench Verified (83.86% at pass@8). Designed to autonomously run tests and iterate until passing results are achieved.
Architecture: GPT-style Decoder
Type: Code
Description: Smaller, faster variant based on o4-mini, designed for Codex CLI and low-latency code Q&A and editing tasks. Default model in Codex CLI. Available on the Responses API at $1.50/1M input tokens and $6.00/1M output tokens with 75% prompt caching discount. Snapshot updated regularly.
Architecture: GPT-style Decoder
Type: Reasoning
Description: Major Codex model upgrade that doubled overall Codex usage following launch. Improved ability to hold large amounts of project data in memory and reason autonomously over extended, multi-file engineering tasks. Described by Codex product lead Thibault Sottiaux as significantly more reliable for autonomous long-running work.
Architecture: GPT-style Decoder
Type: Reasoning
Description: Current flagship Codex model. Tripled weekly active users to 1.6 million at launch. Most capable agentic coding model in the Codex lineup as of early 2026, running 25% faster than GPT-5.2-Codex for Codex users. Deployed by enterprise customers including Cisco, Nvidia, Ramp, and Rakuten.
Architecture: GPT-style Decoder
Total Funding
$11.3B
Rounds
1
Funding Update
N/AMay 2025
OpenAI Codex is not an independently funded company or startup. It is a product line developed and wholly owned by OpenAI, Inc. OpenAI has raised over $57.9 billion in total funding across all rounds (with a $300B+ valuation as of 2025), which funds all OpenAI product development including Codex. There are no standalone Codex funding rounds to report. The Codex CLI is open source (MIT license) on GitHub. The cloud Codex agent runs on OpenAI infrastructure and is monetized through ChatGPT subscription plans (Plus at $20/month, Pro at $200/month, Business, Enterprise, Edu) and via API token billing (codex-mini-latest at $1.50/1M input, $6.00/1M output tokens).
Thibault Sottiaux
Head of Codex — OpenAI
Wojciech Zaremba
Co-Founder & Original Codex Lead — OpenAI
Mark Chen
Chief Research Officer & Early Codex Researcher — OpenAI
No direct competitors available.
2026 · Feb 2
OpenAI launched the Codex desktop app for macOS — a multi-agent command center allowing developers to run parallel coding tasks in isolated agent threads organized by project. The app passed one million downloads in its first week. Integrated with Codex CLI and IDE extension session history. Worktree support allowed multiple agents to operate on the same repository without conflicts.
2026 · Feb 12
GPT-5.3-Codex launched as the most capable Codex model to date, tripling weekly active users to 1.6 million. GPT-5.3-Codex-Spark launched simultaneously as an ultra-fast, lower-latency variant powered by Cerebras hardware for real-time prototyping. OpenAI doubled rate limits across all paid tiers and opened Codex to Free and Go users during a limited promotional window.
2026 · Mar 1
OpenAI launched Codex Security, an application-security agent that builds a threat model of a repository, identifies vulnerabilities, and proposes fixes. In private beta since 2025; tested on 1.2 million commits over 30 days, identifying nearly 800 critical and 10,000+ high-severity issues in Chromium, OpenSSL, PHP, GOGS, and GnuTLS. Also announced acquisition of Python toolmaker Astral to deepen Codex's Python ecosystem integration.
Industry: Software Development
Compliances: Not specified
Integrations: GitHub, VS Code, Codex CLI, OpenAI Responses API, OpenAI Agents SDK, MCP Protocol, ChatGPT, Azure OpenAI, Codex Autofix CI
Support:email, help center, developer docs, enterprise support, community forum
Target audience: Software Engineers, Senior Developers, Engineering Teams, DevOps Engineers, Enterprise Engineering Organizations, Open Source Maintainers, CTOs, AI Builders, Platform Engineers
Supported languages: English, Python, JavaScript, TypeScript, Java, C++, C#, Go, Rust, Ruby, PHP, Swift, Kotlin, Shell, SQL, HTML, CSS, YAML, JSON, Terraform
Astral
March 1, 2026
N/A
0 reviews
No reviews yet
Be the first to share how OpenAI Codex performs for your workflow.
0.0
Accuracy
0.0
Ease of Use
0.0
Output Quality
0.0
Security
0.0
No social feed available for this tool yet.
The original OpenAI Codex was a GPT-3-based code generation model released in 2021 that powered the first version of GitHub Copilot. It was deprecated in March 2023. The current Codex — launched in May 2025 — is an entirely different product: a cloud-based autonomous software engineering agent powered by codex-1 (a version of o3 optimized for coding) and later GPT-5.2 and GPT-5.3 Codex models. It does not just suggest code — it autonomously completes entire tasks in sandboxed cloud environments and opens pull requests for review.
Codex is bundled into ChatGPT subscriptions — there is no separate Codex subscription. It is available to ChatGPT Plus ($20/month), Pro ($200/month), Business, Enterprise, and Edu plan subscribers. For a limited promotional period, OpenAI also opened Codex to Free and Go users. Developers can also access Codex capabilities via the API using the codex-mini-latest model, billed at $1.50 per 1M input tokens and $6.00 per 1M output tokens with a 75% caching discount.
Codex is available across four surfaces: (1) ChatGPT web app — the cloud agent that executes tasks asynchronously in sandbox environments; (2) Codex CLI — an open-source, Rust-built command-line tool for local terminal-based coding; (3) IDE Extension — integrates Codex into VS Code, JetBrains, and Xcode via the Codex SDK; and (4) Codex macOS Desktop App — launched February 2026 as a multi-agent command center with parallel task management. A Windows app was in development as of March 2026.
Copilot and Cursor work in real-time, offering inline autocomplete and suggestions as you type. Codex works asynchronously — you delegate a full task (e.g., 'implement user authentication') and Codex executes it independently in a sandboxed cloud environment: reading files, running tests, and proposing a pull request when done. This delegation model makes Codex better for larger, multi-file tasks but unsuitable for real-time pair programming where immediate feedback matters.
Codex launched in May 2025 powered by codex-1, a version of OpenAI o3 optimized for software engineering via reinforcement learning on real-world coding tasks. The CLI defaulted to codex-mini-latest (a faster o4-mini variant). In December 2025, GPT-5.2-Codex launched, doubling overall usage. In February 2026, GPT-5.3-Codex became the flagship model, tripling weekly active users. GPT-5.3-Codex-Spark — a Cerebras-powered ultra-fast variant — launched in research preview for Pro users the same month.