ollamacloud

[paraflu]

Any ollamacloud model gave me this answer

Error: [ollama-cloud/deepseek-v4-pro] [403]: this model requires a subscription, upgrade for access: https://ollama.com/upgrade (ref: 41bfac68-523a-4052-8f6f-cc978cc92a26) (reset after 1s)]

But deepseek is an available model... https://ollama.com/library/deepseek-v4-pro

[GravityGremlin]

Some models, including Deepseek-v4 are only available for subscribers, others like Gemma-4 are available for free.

[diegosouzapw]

Thanks @GravityGremlin -- that's exactly right. The 403 is Ollama Cloud's own subscription gate (the message and the ollama.com/upgrade URL come straight from upstream), not an OmniRoute error. deepseek-v4-pro is a paid-tier model; free-tier ones like gemma4:31b work without a subscription.

That said, @paraflu -- your "any ollamacloud model" wording pointed me at a real bug on our side, and I confirmed it in the code. When a paid model returns that subscription 403, OmniRoute currently cools down the whole ollama-cloud connection instead of locking out just the paid model. So right after you hit deepseek-v4-pro, the free models on the same key get temporarily knocked out too -- which is almost certainly why it looked like every model failed.

Root cause: ollama-cloud is a passthrough provider (many models behind one key), so a per-model failure should stay model-scoped. But the subscription 403 slips past every model-lockout path and falls through to a connection-wide cooldown (src/sse/services/auth.ts only model-locks 404/429/5xx, and the one 403 model-lockout case is hard-coded to grok-web). I've opened #3027 to fix it -- a model/permission 403 on a passthrough provider should lock only that model and leave the rest of the connection serving, matching how grok-web already behaves.

Until the fix ships: build your combo from free-tier Ollama Cloud models (e.g. gemma4:31b) and keep paid-tier models (deepseek-v4-pro, kimi-k2.6, ...) out of it unless your Ollama account has the matching subscription. If you do hit a paid model, the connection self-recovers after the short cooldown shown in the error (reset after Ns).

Will update here when #3027 lands.

[diegosouzapw]

Closing for inactivity -- the 403 is Ollama Cloud's own subscription gate (the response body and the upgrade URL come directly from Ollama's servers, not OmniRoute). Feel free to reopen if behavior changes, or open a fresh thread with a current log. Thanks!