Codex for open source | Hacker News Reader

Codex for open source

openai.com

275 pointsEvgeniyZh118 comments

Same as Anthropic's similar offer, they're only giving 6 months free. This really just feels like a way to get OSS maintainers hooked so they buy subscriptions after the free period is over.

If they were really serious about supporting OSS, they'd offer it for free perpetually (well, with periodic checks to ensure the maintainers are still affiliated with the project). Anything less just makes it look like a marketing stunt.

And also, dumb how Github-centric this is, same as Anthropic's signup form. Most of my OSS contributions aren't on Github. Guess that means the projects I've worked on don't matter.

Presumably it's also sessions that they will absolutely use as training data?

The bulk of open source development is (still?) on Github.

It's not like the Linux kernel isn't real. It's just that the kind of people who write Linux kernel patches and get them accepted are, in the eyes of an average open source developer, somewhere between "majestic magical creatures" and "madmen".

is there any company or product that donates their subscription service free forever ?

Github?

I read anthropic's offer is majorly nerfed compared to their regular 20x plan. I've seen OSS develoeprs ask anthropic to revert their plan to a paid plan because it's otherwise a hindrance.

Seems rather stingy - 6 months is barely longer than you will get on a free signup deal for a lot of online products anyway. Kind of worse than nothing if it causes you to adopt work patterns that aren't sustainable for the project after the offer ends.

Which online product gives away 6 months of a $100 per month subscription?

That's what I was thinking, but the downvoters are hunting today.

6 email addresses gives you 6 one months trials …

Where am I going to find multiple+email@adress.es?

One can consult the oracle, /dev/random

Using a plus sign is subaddressing [1] and most ESPs[2] will route to the main address ( multiple@addre.es) . So you can use use multiple+email@adress.es, multiple+xyz@adress.es and both will route the email to you.

In my experience most SaaS apps do not filter this out and allow re-sign ups with sub-addresses.

Gmail has an additional behavior that dot character is ignored in local component of the address . multiple@gmail.com, mult.iple@gmail.com mult.ip.le@gmail.com all route to the same inbox as well.

[1] https://datatracker.ietf.org/doc/html/rfc5233 [2] Less common in work hosted ESPs but almost universally default enabled in public ESPs for consumers.

This is not true (anymore?). I have a rather unfortunate exact naming collision with a family member. They use the full name without dot for the local gmail component, I use a dot between the first and last name.

Two or three mails have been misplaced in a decade.

It is still the rules for Gmail (https://support.google.com/mail/answer/7436150?hl=en).

It would be feasible to change something like that without breaking security now.

Google can hardly start allowing/routing a new account for first.last@gmail.com when you were getting it for years even though your account is firstlast@gmail.com and sensitive communication like say from your bank would routed there.

icloud+ hide my email for $0.99

you need a non-voip phone number for codex SMS now, as well.

This. I tried even tried signing up with my paid business line and it denied me due to VoIP.

Since when did they have trials?

I’m running on my 3rd codex trial and have had a month of Gemini Pro. I think Claude is the only one without trials.

> OpenAI may reject, suspend, or revoke any Program benefit for any reason in its sole discretion, including without limitation if it reasonably believes that an applicant or recipient: [...] (ii) used multiple identities or accounts to obtain more than one benefit

Every 2nd SaaS with a startup plan? I used intercom/customer.io/segment/amplitude/mixpanel for free for a year.

The marginal cost of all of those is definitely much lower than for Codex, though

The price might be more commoditized if OpenAI kept true to the original mission that lives on, albeit vestigially, in their name.

Make it $200/month subscription which actually gives you access to O($1K) worth of codex compute. Even at the face value it is very generous, IMO.

An online product that was brought into existence by processing all the open source software in the world and makes money by selling the resulting knowledge base, should be accessible free of charge by the producers of that open source software.

Jetbrains gives away for free infinity years of a $180+ per year subscription (its more expensive in the first year or for orgs)[1] for open source authors, students, and more. Sure, the per-month price tag is not as high but after year 4 you saved much more.

[1] https://www.jetbrains.com/store/?section=students&billing=ye...

Though it doesn't cost them anything of note to provide this other than maybe some lost sales to devs who would have bought it.

After using JetBrains IDEs for years I can hardly really get into anything that isn't vertically integrated. Language servers are THE WORST -- I love Zed but only use it for things that don't require language integration at all.

It's like how after using Apple hardware for years I couldn't put up with most Windows laptops -- either they were HiDPI ultrabooks with no performance or they were sloppy gamer machines with no class.

Learning JetBrains gets you hooked.

No good deed goes unpunished

They're doing everything possible to drive up their MAU before their IPO.

FYI this program is ~3 months old, and Anthropic has a similar Claude for Open Source program (see https://claude.com/contact-sales/claude-for-oss).

I applied for both. Heard back from neither. Mentioned two particular projects when applying, one with 2k stars and 5M monthly downloads, and another with 2M monthly downloads.

Now I'm wondering what the bar is since even people with millions of users aren't making the cut. I'm orders of magnitude smaller but I signed up too since I had nothing to lose. Didn't get a response, of course.

Is it possible this is vapour marketing and no projects are actually being selected? Perhaps someone from a project who has heard back can respond here?

Anthropic released Fable with builtin prompt injection for sabotage purposes. I suppose anything is possible.

I applied for my oss project Filestash which satisfy all their criterias but never heard back despite the project having millions of users, more than 14000 stars on github and representing more work than a single person can cope with

I doubt considerations are based on need. Filestash is cool, but probably isn't the marquee marketing opportunity they are looking for; it jas to be a household name they can name-drop or place a logo on a marketing page and get instant street-cred "${AI_MODEL}: Used* by the React project in 95% of PRs closed last quarter"

I see, so it was never about giving back to the conmunity whose works they used to train the LLMs in the first place...

Not only that, but such sponsorships would also generate new, high-quality end-to-end training data from targeted projects, including work-in-progress edits and prompts which are not present in scraped git histories.

Same here, 15k stars, 150M downloads and never heard back.

Same here ~13k stars. Never hever heard back from them. So frustrating.

Also applied for Codex Open Source - within 2 days I got confirmation and using it since then. Great Job OpenAI! Shame on you Antrophic for not even sending out a refusal message with the cause.

I applied for the first time a couple of months ago and again this month, but unfortunately I haven’t heard back from them :(

I’m building EasyInvoicePDF - a free and open-source invoice generator. (900+ GitHub stars, 2k monthly users on average, 10k total invoices downloaded)

https://github.com/VladSez/easy-invoice-pdf

If you only give 6mo then this is the opposite of a commitment to open source it’s a drug dealers tactic of giving the first taste for free.

> If you only give 6mo then this is the opposite of a commitment to open source it’s a drug dealers tactic of giving the first taste for free.

Its arguably even more self-serving than the drug dealer tactic because of the feedback loop involved (if you use it to maintain your open source project, OpenAI will surely use that new code [along with all the existing code in your project] to train future models).

So it would be like if the drug dealer gave you the first taste for free and also the drug caused you to shit out more drugs and the drug dealer harvested your shit to sell to both future you plus other people.

Very vivid analogy that I'm never going to forget now.

drug dealers should teach MBA classes at this point - so many strategies pioneered by them get used by large tech companies

What does this clause here mean and why would they include it? https://developers.openai.com/codex/codex-for-oss-terms#7-su...

Isn't the thing open source and governed by its own license?

That is interesting. I would have thought they had that right without needing to add it to the ToS.

It's better to have something in writing than to possibly have lawyers argue over it in court.

I did fill the form our a while back (it was around for a few months now) without any response. I guess must be really big OSS project for maintainer to qualify.

What project did you apply for?

PHP

If you harass the right person on Twitter, you could probably get that ball rolling a little faster.

Same. But I got one from Anthropic.

a huge aspect of open source is the user -> contributor -> maintainer pipeline. maybe they mean well, but in fact they're constructing a wall between those last two groups.

especially in larger projects where maintainership duties are heavily delegated, the last thing i want is some tool that can only be used by me, because suddenly i can no longer share the workload that tool targets with people who aren't "technically" maintainers.

6 whole months?! Gee golly thanks mister!

Agreed. Seems like it should be indefinite given they created a multi billion dollar company off the backs of these maintainers dedicating their hard earned timed for free to begin with and then trained models against their code.

IMO this is an insult if anything

Especially considering they trained the damn thing on our code.

Do you complain to the bank that credit cards have expiration dates or to the government that passports also do?

The bank mails me a new one automatically.

"Critical open source software" should not, and maybe cannot, be maintained with its development requiring huge commercial-corporate infrastructure in the form of OpenAI's LLMs.

It should be maintained by humans, relying on widely available hardware and software, requiring little of both.

Not saying that using LLMs as a convenience is forbidden or anything, but the direction is problematic.

(Also, this sounds like a cheap alternative to actually funding FOSS work.)

I think programs like this are cool, the company gets to promote their product and do good at the same time. This looks like a broader program than past ones and giving out GPT5.5 could be meaningful in improving open-source projects' security.

Mycli (https://github.com/dbcli/mycli) is a happy recipient of sponsorship from this program. OpenAI asked for nothing in return; not even a link.

Are you saying they aren't getting training data from you?

I'm sure they are getting training data! But it is hands-off otherwise.

6 months a bummer, but we got it for apple sandbox - coderunner (https://github.com/instavm/coderunner)

We got it yesterday, maybe they just started rolling it out and hence op posted this.

I applied last months ago and again, but there not have any information, but Claude is very fast. I build the https://github.com/go-vgo/robotgo, https://github.com/go-ego/gse and others, 20k+

I wonder how well this supports niche languages. There's an indication there for stars or other signals of importance to 'the ecosystem'; that could match the Big Libraries but likely not ones for small languages.

theprimagen called this[1] like three days ago. That was fast.

[1] https://www.youtube.com/watch?v=m-bT5v5Tm7w&t=164s

They’ve been doing this since at least March

No, he didn't? He predicted that third parties would donate tokens to FOSS projects, not that the labs would. One is PR that started ages ago, the other is a reasonable prediction of where the world is going.

Not quite donate tokens directly (technically and practically weird), but donation -> compute has been out for a couple months on opub.dev (disclaimer, built it). So his prediction was somewhat correct if not late!

Of course the solution to overusing AI is... to use more AI. Love it.

I wouldn't use anything but open weights models for developing open source software. This is just training OpenAI and Anthropic to steal your work for their proprietary models.

The crux of open source: per definition it is opened for the public to use it.

I see it as a chance. Many OS projects themselves offer LLM readable websites, their docs.

This way the project at least not only gets ingested but receives referential treatment.

Some sort of collaboration. Ingested it will be, anyway.

> I see it as a chance

absolutely. AI is the same as any other software, and open source has to integrate, adapt, and lead to make sure that open source values continue to propagate.

my personal approach is to focus on developing with open weights models, so that my work is optimized for them, and leads to their development. proprietary labs are free to copy, but they have a structural cost disadvantage. my objective is that open weights models remain competitive on capability but lead on capability/cost.

If your license is open then Anthropic and OpenAI are using your work anyway.

the user prompts and harness used for development are much more valuable for training than the final source code.

my approach to open source development with AI now is to include all of the agent sessions used in development in the repository, which makes this data freely available for training for both proprietary and open weights models, but that is just my own approach. every open source developer ultimately has to make their own judgement on the best way to integrate AI in accordance with their values.

What are AI companies doing to respect open source licenses and copyright?

I'm sure they train their models on open source software, so how do I know that LLM generated code doesn't reproduce substantial chunks of, for example, GPL licensed code? If indeed there are GPL violations, what are AI companies doing to police themselves?

I wonder if open source licenses will start to include "not to be used for LLM training" clauses.

> I wonder if open source licenses will start to include "not to be used for LLM training" clauses

As if the LLM trainers would care. They've ignored every single license and copyright policy out there because "fair transformative use". It's undergoing litigation in various jurisdictions, and the chaotic side of me really wants to see what happens if a UK or California decide that training an LLM on pirated copyrighted material is not fair use, and the rights holders have to be compensated.

The amount of gift-horse-mouth-looking in this thread is amazing to me.

How dare they only give me this much free stuff! I want that much free stuff!

I want my license fee from my software that they trained their models on.

(Actually I don't, I want their stuff as Free Software and I mean everything, training data, pipelines and all)

How is this different from https://openai.com/form/codex-open-source-fund/ and are the winners listed anywhere? I've only ever seen devs say it isn't worth bothering, many of which I would've expected to be shoe ins for something like this.

The former has been refreshed and rebranded. The new form URL is https://openai.com/form/codex-for-oss/

and both of them generally don't work unless you have bought many github stars

my guess is they get high quality training data.

This is correct. The most valuable form of data for any AI company is corrective feedback from real use cases.

A few AI companies giving free stuff to OSS maintainers - Dosu - CodeRabbit - OpenAI - Anthropic

https://www.oss.fund/explore/?pillar=operational-support&cat...

What a visionary Stallman was.

Applied in March when it first launched for VT Code, a Rust-based terminal coding agent, but haven't heard back from OpenAI. The bar seems high, which makes sense given the fund's limited scope and requirements.

a very good way of collecting high quality training data.

i imagine the usage from maintainers of high quality projects are excellent training data. much better than average joe

The difference between this one (good) and the Anthropic program (bad) is that openai doesn't force you into a marketing clause while Anthropic does.

I mean seriously, you already ripped off all the worlds open source code. Be more generous and don't demand anything else back. Six months is so little too.

Nice way of guaranteeing access to source code as training material and intelligence gathering

I like that a project only qualifies if it's hosted on GitHub.

First hit's free

Hurdles, more hurdles.

Applied twice and didn't get any response

These grant programs feel inconsistent—sometimes they genuinely help OSS, other times they look more like marketing. Hard to tell where the balance really lies.

When in doubt, go with marketing. There are things that are 'just marketing' you wouldn't believe.

> sometimes they genuinely help OSS, other times they look more like marketing

Whenever companies do things like this, it's both, or at least trying hard to be. To the extent that it's perceived by developers (that is, potential OpenAI customers) as helping OSS, it's effective marketing. This perception may or may not correspond to reality.

They are coming for the repositories now.

Codex for open source stored in GitHub*

After what just happened to Anthropic, no way in hell will I ever use, support or give money to Kushner's OpenAI.

That was Amazon's doing: https://news.ycombinator.com/item?id=48519092

Correction: only in part

The Axios article[1] I read says "calls from Amazon — as well as at least five other companies to a variety of senior administration officials Thursday evening and Friday morning — led to the model being shut down by Friday night".

Yes, Amazon is the only company named, but would anyone be surprised if OpenAI was one of the other five companies? It's hard to imagine a company that would materially benefit more from this event.

The evidence is circumstantial, of course, but can you blame people for making a connection?

[1] https://www.axios.com/2026/06/13/anthropic-amazon-white-hous...

it's hard to trust them when there is little human support behind the scenes

just applied

I've forked tensorzero after they archived the repo and will be updating and fixing issues going forward.

https://github.com/agentify-sh/gateway

this is my 2nd attempt

I am using my idle codex usage but would benefit from more inference

what they do when someone switches from open source to closed source later

6 months is a joke.

Companies a thousandth their size are giving free or at-cost access for OSS projects.

"6 months of ChatGPT Pro, which includes Codex" - come on, just make it free. Last time I checked OpenAI was worth $852 billion.

This valuation is derived from the fact the three tokens aren’t free

Trying to get https://opub.dev off the ground to solve this in a more open way.

If you have more than 100 stars, you can get $50 in starter credit.

Ideally organizations, more so than people, provide the bulk of future donations.

As for this program, ehh... Sceptical in general of any frontier program that ends at some time.

Once you're embedded, and all that...

Anyone know how much API credit openai offers?

They give out the subscription by default, and if they find your use case interesting enough they'll give you credits. Not sure if there's an upper limit, but I would be surprised if it's more than a few hundred dollars a month.

(no internal knowledge, this is based on my experience with explainshell.com, thanks OAI!)

On a side note, am I the only one who feels Codex models have a higher general first-pass success rate than Claude models on coding what you want? I use Github Copilot and always find myself drifting more towards them when working.

The moment a corporation starts to endorse open source is the moment they admit they know that are behind.

Anthropic published essentially the same offering recently. By your logic, does that mean they're behind too?

Reply on news.ycombinator.com