No free model is working, 0 models pass on every provider I tried

[shettyshubham72]

Hi, I can't get any free model to work in OmniRoute and I'm not sure what I'm doing wrong.

Setup

• OmniRoute version / install method: [version, Docker or npm]
• OS / browser: Windows 11, Chrome

What I see

• Providers > Free category > "Free Test" says: "0 tested, No active connections found for this group."
• Arena (Free) / lmarena: connection "main" shows connected, but "Test all models" says "0 of 48 models working" (Retest done).
• Ollama Cloud: "0 of 12 models working" (gpt-oss:120b, deepseek-v4-flash, glm-5.1, minimax-m3...). The connection toggle was off at first; I turned it on and retested, same result.
• Adding a connection and pressing Check shows "Invalid API key".
• Using auto/... models from Claude Code gives [403 / 503, paste exact error].

What I tried

• Re-pasted the keys with no spaces and created fresh ones
• Set a Validation Model under Advanced Settings: [yes / no]
• Tested a specific model directly, not auto: [result]

Questions

1. Which free providers currently work without a paid plan?
2. For Arena (Free), what exactly should go in the key field (API key or session cookie)?
3. Is "0 of N models working" a known issue in the current version?

Happy to share app.log lines (keys removed) if needed. Thanks!

[pacocartones]

On question 2: Arena (Free) is a web-cookie provider (internally still lmarena), so the field expects your browser session, not an API key. If you pasted an API key there, that would explain "Invalid API key" on Check. The hint in the code (src/shared/constants/providers/web-cookie.ts, lmarena.authHint) says to paste the full Cookie header from arena.ai: open DevTools → Network, pick a request to arena.ai, and copy its Cookie request header. Include arena-auth-prod-v1.0 / .1 and, when present, cf_clearance / __cf_bm. If Arena still answers 403, it also mentions setting providerSpecificData.recaptchaV3Token from a live browser session.

I can't say from the code alone which other free providers currently work, or why "Free Test" finds no connections. I can't reproduce those two from here, so I can't answer them. The app.log lines you offered, and the exact 403/503 text from Claude Code, would help whoever can. Please remove any cookie, token or key values before posting logs.

[diegosouzapw]

@shettyshubham72 Thanks for the details, and thanks to @pacocartones for the Arena cookie answer. A few things I can add:

• Your post still has the template placeholders ([version, Docker or npm], [paste exact error], [yes / no]). Without the version, how you installed it, and the exact error text, nobody can tell why a provider fails, so please fill those in.
• "0 tested, No active connections found for this group" on the Free Test button means that group has no connection that is enabled for testing, not that the providers are broken. Web-session providers (like Arena) are deliberately skipped by "Test all models", so a "0 of N working" there doesn't mean the models are down. A real request is the only reliable test.
• For Ollama Cloud, "Invalid API key" on Check usually means the key was rejected by Ollama, so please make sure it's an Ollama Cloud key, not a local Ollama setting. I haven't tested it myself.
• I can't give you a verified list of free providers that work today without a paid plan: free tiers change often, and our catalog (docs/reference/FREE_TIERS.md) records when each was last checked, which isn't the same as "works right now".

Next step: send the OmniRoute version, install method, and one failing request from the request log (hide keys), and we can look at one provider at a time.

[rosset]

hi folks, I have the same question, tried some "free" models but no luck, I was unable to find anything that barely works, all looks like a "scam" trying to get your credit card and sold credits to you.

for now I'm using OmniRoute only to route to my network llama-server and to a couple paid services (gemini and chatgpt).

there are some criteria to add the "free" tier models ?

I know there are no such thing as a "free lunch", just want to undertand how others are using the free tier models...

thanks!

[diegosouzapw]

@rosset Fair question. The criteria are written in docs/reference/FREE_TIERS.md (the "Methodology" section): an entry needs a source (a public page, a public endpoint, or a figure someone read in the provider's console, with the date), and each provider has a terms-of-use column (ok / caution / avoid). What that does not mean is "works right now for everyone": free tiers change, some need signup, KYC or a card, and the file records when each was last checked, not that it works today.

I don't have a verified "these N work today" list, so I won't give one. How others use them, from what I can say: the cookie/web-session providers can work but can also get your account restricted, the keyed free tiers (a signup and a key) tend to be the most predictable, and routing a paid or local model first with a free one as a fallback is a common setup. If you tell us which two or three you tried and the exact errors, we can look at those specifically.