Stealth release
The model runs under a food-themed code name. No developer, no model card, no weight release.
Crouton Alpha appeared on OpenRouter in early November 2026 as a free stealth model built for scalable coding, agentic and multimodal workflows. OpenRouter describes it as a highly efficient reasoning model, and third-party write-ups in launch week report 1M tokens of context. The lab behind it has not been named. This page records what is confirmed, how to call it, and which paid models to use when you need a documented API.
Crouton Alpha is a code name, not a product name. It is listed on OpenRouter, the multi-provider router, as a free model aimed at the work agents do repeatedly: coding, tool use, and mixed text-and-image input. That pattern is how several recent frontier models first appeared — anonymous listing, benchmark screenshots in developer communities, then an official name a few weeks later. Until that announcement, every capability claim about the model is provisional, including the context window.
The model runs under a food-themed code name. No developer, no model card, no weight release.
The published listing is free to call. No paid version of this name has been announced.
Access goes through OpenRouter rather than a first-party endpoint, so routing and logging are on the router side.
This is the short list of facts that hold today. Anything not on it — parameter count, training recipe, benchmark scores, the lab behind it — is not public.
Register for TokenRa, open the dashboard, create an API key, confirm that Crouton Alpha is enabled for your account, and then send an OpenAI-compatible request from your server. Keep the key private and verify the live model identifier before production use.
Register, then open the TokenRa dashboard.
Open the API key or token section, create a key, and store it server-side.
Confirm the enabled model ID, test a representative workload, and add production controls.
curl https://tokenra.io/v1/chat/completions -H "Authorization: Bearer YOUR_API_KEY" -H "Content-Type: application/json" -d '{"model":"crouton-alpha","messages":[{"role":"user","content":"YOUR_TEST_PROMPT"}]}'Crouton Alpha is available on TokenRa at $0 per token during the free preview period. Get an API key at tokenra.io/register and use model ID crouton-alpha.
Nobody has said, and that is the honest answer. The model is published under a code name with no developer named, no model card, and no weight release. Launch-week discussion in developer forums has already produced guesses — one write-up argues from serving behaviour and output style that it is a Meta release — but a guess is not evidence.
What you can check for yourself: OpenRouter sometimes lists an upstream provider or a data-policy hint on the model page, and the response headers or billing line can name the provider that actually served your request. Those are first-hand signals. Screenshots in a forum are not. If a name does get published, this page will say which lab it turned out to be and where it is available through TokenRa.
Speculation that Crouton Alpha is a specific lab's model is unverified and changes weekly. Search engines are already full of pages asserting one identity as fact — do not carry that claim into your own product docs or pricing pages.
A free stealth model is fine for evaluating output quality. It is a poor foundation for production: the name changes when the model is unveiled, limits move, and there is no commercial agreement behind it. If you are shipping, route the same request to a model with published rates and a stable identifier.
The other free stealth release from the same week, on OpenCode. Same rules: code name today, maybe a real name later.
The August stealth release, later confirmed as GLM-5.3-Flash. Paid per-token access is available now.
Compare current per-token pricing across the Chinese model APIs TokenRa routes.
Every code name in this table followed the same path: anonymous listing, free or discounted preview, then either an official name or a quiet shutdown. The last row is a reminder that some of them simply disappear.
| Code name | Listed by | Appeared | Access | Official name |
|---|---|---|---|---|
| Crouton Alpha | OpenRouter | 2026-11-01 | Free listing | Undisclosed |
| Pulse Alpha | OpenCode | 2026-11-03 | Free preview | Undisclosed |
| Ox Alpha | Router listing | Earlier preview | Free, now paid | GLM-5.3-Flash (Z.ai) |
| Omen Alpha | Router listing | Earlier preview | See model page | See model page |
| Union Alpha | Router listing | Earlier preview | See model page | See model page |
| Space Bunny Alpha | Router listing | Earlier preview | See model page | See model page |
Dates and status are recorded when each page is updated. Open the linked page for current rates and the disclosed developer. The "listed by" column names the service that published the code name, not the lab that trained it.
Four signals tell you the preview is turning into a real product, and the first one usually shows up in developer communities rather than in an announcement.
If you are evaluating Crouton Alpha for a real workload, record your own results now: same prompts, same scoring, run against the paid model you would use instead. When the name is published, that comparison is the only thing that still has value.
Free previews are usually free because the operator wants evaluation data — and here there are two parties to consider. OpenRouter routes your request, and the unnamed upstream provider generates the completion. With a code-named model there is no model card to check and no data-processing commitment published under that name.
Treat every prompt as readable by both parties: do not send customer records, credentials, regulated data, or anything you would not put in a public issue. If your use case requires a data policy you can point a client at, use a model whose provider publishes one — that is what the paid catalog is for.
Crouton Alpha is a stealth model listed on OpenRouter in early November 2026. It appears as a free listing under a code name, positioned for coding, agentic and multimodal workflows, and OpenRouter describes it as a highly efficient reasoning model. The lab behind it has not been disclosed.
Yes, the listing published so far is free to use. No paid tier has been announced under this code name, and free preview access can change or end without notice.
The duration is not published by OpenRouter. A launch-week community post said the model would be free for about a week, but that is a third-party claim, not an official statement. Check the OpenRouter listing for the current terms.
It is served through OpenRouter's OpenAI-compatible endpoint, https://openrouter.ai/api/v1/chat/completions, and the code name appears in the listing path. Confirm the exact model ID string in the OpenRouter console before shipping code, because a preview listing can be renamed.
Third-party write-ups published in launch week report 1M tokens of context. Verify the figure on the OpenRouter listing itself before you rely on it, since preview specifications are not guaranteed to hold.
That is not disclosed. The model is published under a code name and no developer has been named. Community speculation about which lab is behind it, including guesses that it is a Meta model, is unconfirmed and should not be treated as fact.
No. They are separate stealth releases. Ox Alpha was later confirmed to be GLM-5.3-Flash from Z.ai; Pulse Alpha is a different free preview from the same week whose official name is still undisclosed; Crouton Alpha is a third, distinct listing served through OpenRouter.
Not at the moment. TokenRa is a routing gateway for paid Chinese model APIs and does not currently resell access to this preview. When the official name is published, access will be offered under that name on the TokenRa model catalog.
No. It is a feedback-collection release served through a router to an unnamed provider, so treat prompts and completions as reviewable. Do not send regulated, confidential, or customer data while the model is running under a code name.
Crouton Alpha is a preview code name listed by OpenRouter, not by TokenRa. TokenRa is a routing gateway for model APIs and is not the developer or owner of the models listed here. Official names, rates, and availability can change; provider and related names belong to their respective rights holders.