zebbern/no-cost-ai: a curated index of free AI services, and what its tables actually tell you

What problem the no-cost-ai list solves, and for whom

The README opens with a one-line pitch: "Spend 0$ on Ai-Models!" and describes itself as "A living index for developers, researchers & curious builders". That framing is accurate. This is a directory, not a runtime. The repository contains a README.md, a LICENSE and a .github directory, and the README itself says the hosted version lives at nocostai.vercel.app, where the same tables are rendered as a page.

The problem it addresses is discovery. Free tiers for chat models, image generators and speech tools change constantly, and the information is scattered across provider pages, forum threads and expired blog posts. no-cost-ai collects the links in one place and, more usefully, sorts them by whether you must create an account. The README puts that split at the top: a "No Sign Up Required" block and a "Signup Required" block, each with its own set of topic tables.

The audience is narrow but real. It suits someone who wants to try a model without handing over an email address, or who needs a throwaway image generator for a mockup. It is not aimed at teams that need a contract, a rate-limit guarantee or a support channel. The README never claims otherwise, and the absence of any pricing, SLA or uptime discussion is consistent with that.

How the no-cost-ai index is organised: two blocks, four topics, one limits column

The structure is plain Markdown tables. The no-signup block lists Chat Interfaces (21 links), Media Generation (5), Voice & Music (3) and Developer APIs & Platforms (4). The signup block lists Chat Interfaces (13), Media Generation (12), Voice & Music (4), AI IDEs (2) and Developer APIs & Platforms (17). Each table has three columns: Link, Free Models, Limits.

The Limits column is where the actual information lives, and it is inconsistent by design. Some entries carry hard numbers, such as grok.com at "3 messages every 2 h", chat.mistral.ai at "10 Messages Resets Every 24h", and groq.com at "30 requests / min, 40 000 tokens / min". Others say "Unlimited" or "Unlimited/Unknown". The README does not define what Unlimited means for any given service, and it does not date the entries, so a limit recorded as unlimited may be a snapshot of one provider's policy on one day.

The Free Models column mixes naming conventions. You will see provider model IDs (`gpt-4o`, `deepseek-v3`, `gemini-2.5-flash`) next to internal names (`wbot-2`, `sur-mistral`, `theoldllm`). The README does not explain the internal ones. There is also a Leaderboards section linking to OpenRouter rankings, AlpacaEval, the Open LLM Leaderboard, Chatbot Arena, MTEB, Artificial Analysis, Yupp.ai and LM Arena, which is a reasonable place to check whether a listed model is any good before you spend time on it.

Using a no-cost-ai entry: from the README table to a first request

There is nothing to install. The repository has no package manifest, no build step and no server code. The README points readers to the hosted page at nocostai.vercel.app and to the tables in the README itself. So the first real use is picking an entry and opening it.

For a chat model without an account, the README lists lmarena.ai, g4f.dev, t3.chat and others. For an OpenAI-compatible endpoint, it lists uncloseai.com with the note "Unlimited, OpenAI-compatible API". The README does not give a base URL, a request format or an example call for that endpoint, so the practical first step is to open uncloseai.com and read its own documentation before writing any client code. The same applies to g4f.dev, whose README entry points to a separate page at g4f.dev/docs/ready_to_use.html and notes that an API key page exists at g4f.dev/api_key.html.

For a no-signup image generator, the README lists pollinations.ai and perchance.org/wtr90dexsn, the latter noting it "Uses `pollinations` api". Both are browser-first and the README gives no request format for either. The only command-shaped material anywhere in the README is the ollama.com entry, which the table describes as "Unlimited run on cloud or local machine"; the README does not print an install command for it, so the honest next step is to follow the link rather than guess a flag.

Where the no-cost-ai index breaks down

The largest limitation is that a link list cannot stay correct. The README calls itself "a living index" and invites contributions, and the last push to the repository was on 2026-09-07. But the entries themselves are undated. When grok.com is listed at "3 messages every 2 h", there is no way to tell from the README whether that was true last week or last year. A provider can change a free tier without the table changing.

Second, free services that front paid models are unstable by nature. Several entries describe access to models that normally cost money, and the README itself flags sharedchat.cn as "`Unlimited` but may be hard to get a responce as many others use it aswell but it really does work". That is an honest note, and it also describes the general failure mode: shared capacity, queueing and sudden removal of access. Anything you build on these services can stop working between deploys with no notice.

Third, there is no reliability or privacy information. The README does not say who operates each service, where prompts are stored, or whether inputs are used for training. For a hobby prompt that is fine. For source code, customer data or anything under an NDA, a free proxy endpoint is the wrong tool, and the index gives you no way to evaluate that risk.

Finally, the AI IDEs table in the signup section is empty. The heading is present with no rows, which suggests the category was planned and not filled.

no-cost-ai compared with running Ollama locally

The README lists ollama.com in the no-signup Developer APIs & Platforms table with the note "Unlimited run on cloud or local machine", and it also links github/ollamafreeapi. That is the closest thing to an alternative approach inside the same document, and the difference is worth stating plainly.

A hosted free endpoint gives you a large model with no download, no GPU and no disk cost, and in exchange you accept someone else's rate limits, someone else's logging, and the chance that the endpoint disappears. Ollama inverts all of that. You download weights, you run them on your own hardware, and the limits are whatever your machine can do. Model quality is generally lower than a frontier hosted model, and the first run costs a multi-gigabyte download.

The choice follows from what you are doing. Prototyping a prompt, generating a throwaway image or testing an unfamiliar model: the hosted entries in this index are faster to try. Running a batch job overnight, handling anything sensitive, or needing the same model version to behave identically next month: local execution is the only one of the two that you control. The README does not make this argument, but its own tables support it, since the local option appears under "Unlimited" while the hosted ones mostly carry message or token caps.

Licence, maintenance and the cost of keeping the list current

The repository is MIT-licensed, with the LICENSE file at the root. That covers the index itself: the README text and the tables. It does not cover any of the linked services, and the README makes no claim about their terms. If you copy table content into your own project, MIT requires you to keep the copyright and permission notice. If you call a listed API, the licence of this repository is irrelevant to that call; the provider's terms apply, and the README does not summarise them.

On maintenance, the last push was on 2026-09-07, so the repository has been touched recently. There are no releases, which is expected for a documentation-only project. The upgrade cost is therefore not a version bump but a re-read: because entries are undated and limits are free text, keeping a fork accurate means periodically opening each service and checking its current allowance. For a list of this size that is real work, and it is the main reason a fork tends to drift out of date. There is no changelog in the repository, so the README itself is the only record of what changed.

Editorial conclusion

Adopt it as a starting shortlist when you need a no-signup chat or image service today, and read the Limits column before you build anything on top of a listed service. Skip it if you need an API with a written SLA, a stable model version, or a permanent free tier, because the index records what a service allows at the moment it was added, not a contract. Verify first: the licence file at the repository root, the current limits on the two or three services you actually plan to call, and whether the model names in the table still match what the provider serves.

Frequently asked questions

Official sources

1. Issues
2. License: MIT
3. Project website
4. README
5. zebbern/no-cost-ai on GitHub