Changelog
New features, improvements, and fixes, newest first.
Nov 14
Saturday 2026
- The AI agent access page is simpler: make a token on one side, copy the one-line setup on the other, with every other way to connect in one section below.
- The Sell page now shows that your AI agent can apply, import, and price your models for you.
- One-time setup codes are longer, in three groups of four.
- Your agent checks your endpoint and key before it applies, and tells you plainly if the key is refused.
- Price changes made through your agent work: cuts apply at once, rises wait 24 hours, as on the dashboard.
- While your application is in review, your agent says so instead of reporting a missing model.
Nov 13
Friday 2026
- The chat site lists only the models your account can use.
- The status page shows the three most recent updates, and its uptime bars fit narrow screens with fewer, wider days.
- The Sell page and the other docs pages use the full width on large screens.
- Paid models have their own section: tokens and requests per day, and the most used paid models.
- Daily usage: tap Tokens or Requests to see one line with real numbers.
- Tokens per day by model shows the top five models, and clicking a model keeps it highlighted. GoonPAI now appears among the top clients.
- Agent connections now stay open through slow steps, such as importing a long list of models.
- Security hardening.
- Sellers: after you make a token on AI agent access, copy one line and paste it into Claude Code, Codex, Cursor, or another agent. The line carries a one-time code instead of your token, so the token never passes through the chat.
- The AI agent access page is redesigned, with example requests to try once your agent is connected.
- Every model now has its own page at freetheai.org/models/<name>, listing every provider's price, speed, and reliability side by side.
- Sellers: each model on your listings page has a Changes button. It shows what changed and when, and whether you, your AI agent, or our staff made the change.
- The Sell page, How it works, the setup guides, and AI agent access now have an "Onboard your agent" button. It copies one line for Claude Code, Codex, Cursor, or OpenCode, and that line sets the agent up for your account.
- Selling on FreeTheAI is open to everyone. No invite or beta code is needed: connect your endpoint, set your prices, and get paid in crypto. Everything about it is on the new Sell page: freetheai.org/sell.
- New: AI agent access. Make a token under Sell > AI agent access, paste the setup prompt into Claude Code, Codex, Cursor, or any MCP client, and your agent can apply, import your models, price them, and submit them for you. It asks before anything buyers can see, and endpoint key and payout changes need your authenticator code.
- Every change to a listing is now kept in its change history, whether you, your agent, or our staff made it.
- New page: how the marketplace works, from adding funds in crypto to how a request picks its provider: freetheai.org/docs/how-it-works.
- Paid models are served by independent sellers. The Privacy Policy and Terms now say plainly that FreeTheAI cannot promise zero data retention for what a seller keeps. Sellers still never get your account, email, or API key.
Nov 10
Tuesday 2026
- FreeTheAI started with a simple idea: give people good AI models for free. Thanks to all of you it grew far beyond anything we expected, and demand is now well past what the free capacity can handle. As of today the free tier is closed for most accounts, and a small number of accounts keep free access for now.
- We are sorry for the disappointment this brings, and we know a lot of you relied on it. Your account, API keys, and wallet balance stay as they are, and paid models keep working through the same API.
- Will it come back? Possibly. If we find a way to run a free tier that works well and is fair to everyone, we will bring it back, and you will hear it on our Discord first.
- The site changed with it: FreeTheAI's own free models are no longer listed, the home page shows live prices, and a banner at the top of the site says what changed. The full note is on the blog.
Nov 5
Thursday 2026
- Prompt pages no longer break when you confirm your age or open another prompt while the site is busy.
- Going back or moving quickly between pages, such as from a model to the models list, no longer shows a service unavailable page.
Nov 2
Monday 2026
- The stats page now shows how many accounts are stopped for breaking the free-tier rules, and how many of those were stopped in the last 30 days.
Nov 1
Sunday 2026
- Providers can now offer embedding, text to speech, speech to text, image, and video models. They are paid models, priced per million tokens, per million characters, per minute, per image, or per second, and you are charged only for what you receive.
- The models page has a Kind filter, and each of these models shows its price in its own unit, such as $0.04 per image.
- Your logs show what these requests were billed for, such as 3 images or 90 seconds of audio.
- The docs have new pages for embeddings, speech, transcription, images, and videos.
Oct 31
Saturday 2026
- Sign-in: if your authenticator app stops working, you can now finish signing in with one of your recovery codes.
- Replies now always report how many tokens they used, streamed replies included, so apps that track your context size see real numbers instead of zero.
- Your logs now show reasoning for models that think without reporting it, marked as an estimate. Charges are unchanged.
- A request too large for the free tier is no longer resent over and over by apps that retry on their own. Lower your context size or start a new chat.
- Stats: point at one model in the daily charts to see it across every day. New: people per day by model, weekly retention, cache hits, prices next to the maker's own, and how each number is counted.
- Models: each model shows the lowest paid price, and you can filter by reasoning, tool calling, image input, and context length. On phones the filters open from one slim bar.
- Provider pages: sort the model table by any column.
- Setup guides: screenshots sit in the middle and open full screen.
- Model requests: vote right from the list, with maker logos and a search.
- Team: the story is rewritten, with live community numbers.
- Dashboard: your API details, newest key, and an example call come first, and the figures are all the same size. Sellers get tabs and one line per model on the listings page.
Oct 30
Friday 2026
- The FAQ, blog, setup guides, docs, changelog, model requests, team, status, and legal pages now use the full width of your screen.
- Setup guides: search for your app, filter by kind, see each app's icon, and copy a test call that checks your key works.
- App icons now show clearly in both light and dark mode.
- Provider pages show every model's figures without scrolling sideways.
- Model pages mark the lowest price, the fastest, and the most reliable provider, and show how many requests each figure is based on.
- Open any provider on a model page to see its reply time, cache price and hits, errors by kind, and last error.
- Turn a provider on or off for your keys right from the model page, and copy ready-to-run code with the Quickstart button.
- When only one provider serves a model, the page says so.
- Your logs show when a request was retried on another provider, which provider answered, and why it was picked. Only the try that answered is charged.
- My providers now shows how each of your keys picks a provider.
- A provider's success rate and error counts now include only requests that reached the provider and failed on its side, not requests we turned away first or ones the buyer stopped.
- A provider model is paused only after it fails two checks in a row, and never because of a problem on our side.
- A new API key from a seller brings its paused models back within a minute.
- Sellers now choose whether we send automatic test requests to their models. With it off, we only call their API for real requests, when they click Check now, and when they update their key.
- Every provider's offer now has its own model ID, like WhiteSummit/gpt-6.1-terra. Requests with it go only to that provider; the plain model ID still lets your key pick the cheapest provider you follow.
- Each provider also has a short 4-letter code that works the same way. You can find both on the provider's page, and the API's model list includes them.
- The models list now shows how many providers serve each model and its largest max output.
- Sellers can check a model again from their listings page with Check now, so a fixed model comes back without waiting.
- Endpoint health on the seller pages now counts only requests that reached that seller. It counted other people's requests for the same model that never left FreeTheAI, which showed errors that weren't there.
- Models from providers whose address uses its own API version could not be reached and were paused. They work again.
- On the Models page, a model offered by several providers now lists all of them together on its page, so you can compare prices side by side.
- New models are filed under their maker on their own, without waiting for an update.
- A new Uncensored filter on the Models page lists every uncensored model.
- Provider pages now show each model's input and output price and context size, and say N/A where there is no data yet.
- Each provider on a model page links to its provider page, where you can follow it.
- Model and provider pages now show how fast each provider starts replying, how much of your input it serves from its cache, and when it last had an error and what kind.
- Sellers can replace their API key from the listings page; the new key is checked before it is saved.
- Seller setup now reads the model lists of many more providers, including Mistral, and one unusual model no longer stops the whole list from importing.
- Sellers can price all their models in one step as a share of the official prices, such as 10% of them, and see what buyers pay and what they earn before sending.
- Paid requests can run three at a time, separate from the free request limit.
- You are never charged for more of a reply than you received or than your request allowed, even if a provider reports more.
- You can now unfollow a provider while your keys use all the providers you follow.
- Asking for a longer reply than a model allows now says so, instead of saying the model was not found.
- FreeTheAI stays at the top of the providers list in every sort.
- API key cards are easier to read, and a new key is shown in full.
- Keys now use only the providers you follow, pick the cheapest one for each request by default, and skip providers that have been failing lately.
- If the cheapest provider is busy, a key moves to the next one only within the step up you allow (25% more by default). Otherwise you get a clear message saying how much more it would cost.
- Free models are used only when you are out of credit, unless you choose to use free models first.
- If a provider fails before replying, your request is tried on another provider you follow, and you only pay for the one that answers.
- Creating a key is now a short three-step guide with explanations for every setting.
- The providers page now shows each provider on one compact row with its models, price, uptime, and rating, so it stays easy to compare as more providers join.
- When several providers serve the same model, the cheapest one is now picked by what your request would really cost, including small requests.
Oct 29
Thursday 2026
- Status bars now show each day as a column: a healthy day is a full green bar, a worse day is shorter and amber or red, and a day without data is grey. A key under each section explains the colors, and each service shows its state right now.
- Uptime figures now round down, so a day with any failures never shows as 100%.
- The Stats charts now appear complete and at full width as soon as the page opens.
- The leaderboard methodology is easier to read, and the maker chart now lives only on Stats.
- Fetching models from your endpoint now works with many more API formats, including lists that span several pages.
- When a fetch fails, the seller page now says why: the key was refused, the address has no models list, the endpoint was busy or unreachable, or the list could not be read.
- Every page shows up sooner, especially on phones, and the home page artwork loads much lighter.
- The home page headline now uses the full width of a phone screen.
- API keys can set the most they pay per million tokens and block providers you do not want. Find it under "Price limit and blocked providers" when you create or edit a key.
- On a model page, each provider now shows its own success rate, reply time, and speed.
- The users-by-country map and the active users figure load reliably again after updates.
- Tavo is now counted as its own app in Top clients, with its full stats.
- The Stats link stays in the top bar on smaller laptop screens.
- The Stats map now colors each country by its users, and model makers and top apps are compared on radar charts.
- Drag across the daily usage charts on Stats and your Usage page to zoom into a stretch of days.
- Each model page shows how its providers compare on success, speed, reply time, context, and weekly use, with success as a gauge.
- Models from OpenRouter, Kilo, Liquid, Poolside, Dots, and OpenCode now show their own logos.