Codex for OSS Offers Maintainers Six Months of Pro

OpenAI runs a programme that gives open source maintainers six months of ChatGPT Pro with Codex, conditional access to its security tooling, and some API credits. It is open for applications right now, on a rolling basis. The offer is real, the form accepts submissions, and the terms are published.
What OpenAI has not published is almost everything else a person would want before applying. There is no launch announcement with a date. There is no statement of how many maintainers have been accepted. There is no dollar value on the credits. There is no published eligibility threshold, not a star count, not a download figure, not a dependent-repository number. And every benefit is framed in the legal terms as discretionary, available if OpenAI decides to provide it, in whatever scope and duration it chooses.
That gap is the story. Codex for OSS is a genuine contribution to the people who keep open source running, and it is also an unusually opaque one for a company that publishes detailed documentation on nearly everything else. This article sets out what the programme actually promises, what the terms let OpenAI do, what the application demands of an applicant, what the offer is worth at list price, how the Codex product behind it is split between open and closed components, and how the whole thing compares with a rival programme that publishes the numbers OpenAI withholds.
What the programme actually promises
The offer has three parts, and the wording shifts depending on which official page you read.
The application page states that selected maintainers receive six months of ChatGPT Pro, which includes Codex, conditional access to Codex Security, and API credits for coding, maintainer automation, release workflows and core open source work. A second block on the same page softens this to maintainers who may receive API credits, with qualified repositories possibly considered for conditional Codex Security access. The move from receive to may receive is not cosmetic, and the terms settle which reading applies.
The developer-facing page adds context the form does not. It describes Codex as supporting maintainers rather than replacing their judgement, and it makes one commitment that deserves more attention than it has received: developers should code in the tools they prefer, whether that is Codex, OpenCode, Cline, pi, OpenClaw or something else. That sentence is the clearest statement of intent in the entire record, because it means the programme is not conditional on using Codex exclusively.
The same page also frames the 2026 benefits as coming out of the earlier funding vehicle, saying the fund now supports eligible maintainers by offering six months of ChatGPT Pro with Codex and conditional Codex Security access for core maintainers with write access. OpenAI does not explain whether the earlier programme and this one share a budget, whether applying to one affects the other, or whether the 2025 grant scheme is still operating. Both application forms remain online.
Every benefit is discretionary
The programme terms are the most precise official text about Codex for OSS, and they are almost entirely composed of discretion.
Approved applicants may receive one or more of the listed benefits, as determined by OpenAI in its sole discretion, and the availability, duration, scope and timing of any benefit may vary by applicant, repository or use case. Submitting an application does not guarantee selection, funding or access. Selection may consider repository usage, ecosystem importance, evidence of active maintenance, role or permissions, and programme capacity. OpenAI may request additional information to verify identity, repository affiliation and maintainer status, and its decisions are final.
The remaining clauses follow the same logic. Benefits are personal, limited, non-transferable and have no cash value. They may expire if not redeemed or activated within a period OpenAI specifies without stating what that period is. Codex Security access and credits are described as optional additional benefits that may require separate review and additional terms, and access may be limited to repositories the applicant owns, maintains or is authorised to administer. Applicants may not use the programme to scan or review repositories they do not own or lack permission to review. OpenAI may modify, pause, limit or discontinue the programme, its criteria or any benefit at any time. Recipients handle their own taxes. The programme is void where prohibited.
None of that is unusual for a corporate programme, and much of it is standard legal caution. It does mean, though, that a maintainer planning around this should treat it as a possible benefit rather than a commitment. The terms also warn that applicants should not submit confidential information, because OpenAI has no duty to treat application materials as confidential.
What it asks of an applicant
The form is short and specific, and a few fields are worth knowing about before starting.
The GitHub profile and the repository must both be public. A ChatGPT account is mandatory, and the email entered must be the one associated with that account. The applicant must state whether they are a primary or core maintainer, choosing between those two labels. They must explain why the repository qualifies in 500 characters, with prompt examples including stars, monthly downloads and ecosystem importance. They must supply an OpenAI Organization ID, which is an unusual requirement for an individual maintainer and is not explained on the page. Two optional checkboxes cover interest in Codex Security and in API credits, and a second 500-character field asks how the credits would be used.
One inconsistency is worth flagging. The page states that developers can apply for their own project or nominate another maintainer, and the documentation repeats the nomination framing. The form itself exposes no nomination path. Every field concerns the applicant and their own repository. Until a route appears, treat the nomination claim as unverified.
The selection signals OpenAI describes are qualitative. It looks for meaningful usage, broad adoption or clear importance to the ecosystem, and it reviews repository usage, ecosystem importance and evidence of active maintenance, including pull request review, issue triage and release management. That is a reasonable description of what maintainership involves. It is not a threshold anyone can evaluate themselves against.
What the offer is worth, at list price
OpenAI does not say which tier of ChatGPT Pro a recipient receives, which makes any valuation a range rather than a number.
The published plans put Pro at from 100 dollars a month, with three usage tiers named in the pricing footnote as Pro 100, Pro 200 and Pro 500, carrying 100, 200 or 500 dollars a month in API credits respectively. Six months at the entry tier is 600 dollars at list price. At the top tier it would be 3,000 dollars.
Those figures are the list value of the subscription, not the value of the grant, and the difference matters. OpenAI has published no figure for individual credit awards under this programme. The only dollar amounts in the public record belong to the earlier fund, which was described as a million-dollar initiative with grants of up to 25,000 dollars in API credits.
For context on what the subscription buys, the Codex pricing documentation lists usage bands that vary by model and tier. At the mid Pro tier the published range runs from 75 to 400 local messages per five hours on one model, up to 300 to 1,750 on a smaller one, all inside a shared five-hour window that may also carry weekly caps. A maintainer trying to size the benefit should read those bands rather than the headline price, because messages per window is the constraint that actually bites during a release cycle.
The Codex you get, and the parts that are closed
The programme is named after a product that is only partly open source, which is worth knowing for anyone whose interest in Codex for OSS is partly philosophical.
According to OpenAI's own documentation, the open source components are the command line interface, the software development kit, the app server, Skills, and the universal cloud environment. The editor extension and the web interface are not open source. The product page describes Codex as usable across ChatGPT, an editor and a terminal, connected by a ChatGPT account, with cloud tasks that continue after a laptop closes and automated pull request review.
Model naming is a trap in current OpenAI documentation. The Codex models page lists four recommended models and describes one as the newest frontier model for complex coding, while the changelog records that same model retiring from ChatGPT and Codex in October 2026, and other release notes describe newer families rolling out. The defensible statement for a reader is that Codex runs on a mix of GPT-5-era and GPT-6-era models depending on plan and surface, and that OpenAI's own pages have not been reconciled. Any article naming a single current model is overstating what the documentation supports.
Codex Security is the third benefit and the most conditional. OpenAI describes it as a plugin or cloud service that finds and remediates vulnerabilities, can scan a repository or a folder, review changes before merge, triage a backlog and propose bounded patches. The cloud version is in research preview. Access is listed for Enterprise, Edu, Business and Pro users, which means the programme is granting a paid-plan feature to maintainers, subject to separate review.
How it compares with the rival programme
The most useful comparison is with Anthropic, because the headline is identical and the disclosure is not.
Anthropic's Claude for Open Source programme offers six months of Claude Ultra at no cost. It publishes five numeric eligibility tests: 500 or more dependent repositories, 100 or more dependent packages, or 200,000 or more combined monthly downloads across registries; listing as a committer or maintainer on a recognised foundation or language project such as CPython, the Rust team, the Node.js technical steering committee, an Apache project, CNCF, Kubernetes, the Linux kernel, Django or Rails; 100 or more pull requests merged into repositories the applicant does not own in the last twelve months; 20 or more unique external contributors with merged pull requests in the last twelve months; or an OpenSSF criticality score of 0.4 or above.
Anthropic also states exactly what happens at the end. The complimentary subscription ends, the company emails beforehand, and an account either resumes a prior paid plan at its previous rate or returns to the free tier. It keeps an escape hatch for people who do not fit the tests but maintain something the ecosystem quietly depends on.
Set the two side by side and the contrast is clean. Same length, same spirit, same invitation to apply anyway. One vendor tells a maintainer whether they probably qualify and what happens on day 181. The other tells them the decision is at its sole discretion and publishes no threshold at all. Anthropic offers one thing. OpenAI offers three, and two of them are conditional.
Whether that matters depends on what a maintainer wants. Someone who just wants to know if they will get the subscription will find the Codex for OSS page frustrating. Someone who wants the widest possible set of benefits, including security scanning that Anthropic does not bundle, has a reason to apply anyway.
The fund this grew out of
Codex for OSS did not appear from nowhere. It extends the Codex Open Source Fund, a million-dollar initiative announced the previous year to support open source projects using the Codex command line interface and OpenAI models, with applications reviewed on an ongoing basis and grants of up to 25,000 dollars in API credits.
That page carried no eligibility criteria, no selection criteria and no recipient count, a pattern the newer programme repeats. It also carried a disclaimer under which applicants acknowledge that OpenAI may develop or receive similar ideas and assumes no obligation of exclusivity.
The relationship between the two is genuinely unclear. The newer page presents the benefits as an evolution of the fund rather than a replacement, yet both forms are live, and OpenAI has not said whether the million dollars is the budget for the subscriptions and credits, or whether the grant programme is closed. For a maintainer deciding where to spend an afternoon on applications, that ambiguity is a real cost.
What this is not
It is worth being precise about the framing, because it is easy to overstate.
OpenAI does not describe Codex for OSS as supply chain security. It does not mention critical infrastructure as a policy category, does not reference incidents of the kind that made maintainer funding a public issue, and makes no national security argument. The company describes the programme as supporting the maintainers behind critical open source software, reducing coding and review load, and preserving security and code quality across widely used projects. The security element is a scanning tool, offered conditionally.
It is also not an announcement in the ordinary sense. Reviews of the record find no launch blog post, no press release and no changelog entry for the programme on any OpenAI property. The earliest archive capture of the application page is 6 March 2026, behind a redirect, resolving fully the next day, and an OpenAI developer blog post dated 9 March links to the programme as an existing thing. Early March is therefore the defensible launch window, and no more precise claim should be made.
For readers thinking about how maintainers get supported across the industry, our look at OpenAI's approach to persistent agent security covers the company's posture on a different problem, and our piece on self-hosted AI coworkers you can audit looks at the open alternative to depending on a vendor programme at all.
There is a practical question underneath all of this that the programme does not answer. A maintainer who accepts six months of Codex access is adopting an agent into a workflow that other people depend on, and the evidence on how coding agents behave in a shared workspace is still thin. Our coverage of how agent workspaces are actually organised looks at the operational side, which matters more to a maintainer than the list price of the subscription. The value of six months of Codex depends entirely on whether the time saved reviewing pull requests exceeds the time spent supervising an agent that writes them.
Questions readers are asking
What does Codex for OSS give a maintainer? Six months of ChatGPT Pro, which includes Codex, conditional access to Codex Security, and API credits. The application page states these as benefits received, the terms state them as discretionary, and the terms govern.
Is it open right now? Yes. Applications are reviewed on a rolling basis with no published deadline, and selected applicants are notified by email.
What are the eligibility requirements? OpenAI publishes none numerically. It looks for projects with meaningful usage, broad adoption or clear importance to the ecosystem, and reviews repository usage, ecosystem importance and evidence of active maintenance.
How much are the API credits worth? Not disclosed for this programme. The earlier Codex Open Source Fund described grants of up to 25,000 dollars in API credits, but OpenAI has not said whether that cap applies now.
Which ChatGPT Pro tier do recipients get? Not disclosed. Pro lists three tiers at 100, 200 and 500 dollars a month, so six months is worth between 600 and 3,000 dollars at list price depending on the tier.
Is Codex open source? Partly. The command line interface, the software development kit, the app server, Skills and the universal cloud environment are open source. The editor extension and the web interface are not.
When did the programme launch? OpenAI published no dated announcement. The application page was first archived in early March 2026 and an OpenAI developer blog post dated 9 March 2026 links to it, which supports early March as the launch window.
Register for OpenAI Codex for OSS here
Sources
- OpenAI developer documentation, Codex for OSS
- OpenAI, Codex for Open Source Program Terms
- OpenAI documentation, which parts of Codex are open source
- OpenAI, ChatGPT plans and pricing
- OpenAI, Codex Open Source Fund application page
- Anthropic, Claude for Open Source programme