Unlimited from $5. Or tokens at cost, forever.

No token meter and no monthly quota on the models in your band — you pay the flat price whatever you consume. What is bounded is concurrency, not volume: a per-minute request rate limit and a concurrent-stream limit per plan (20/1 on Lite, 60/3 on Standard, 120/5 on Max), personal single-person use, and no bulk generation of training data or resale of access.

Lite covers hosted open weights under 50B parameters, such as Llama 3.3 8B, Qwen3 32B and Gemma 27B. Standard adds everything under 200B, including DeepSeek V3.1 and Qwen3 235B MoE class builds where the active parameter count sits in band. Max covers the whole hosted fleet with no cap. Every model page states its band.

No — it extends it. You keep paying the lab directly for the model you like; the $20 you pay us is applied toward that frontier subscription cost, and what you get from llmcloud is unlimited open-weight coverage the moment the lab's daily limit stops you.

We detect the rate-limit response and re-route that request to the best open-weight model on our own GPUs for that workload, in the same session and through the same API call. It is free — no token charge, no overage. When the lab's window resets you go straight back to the frontier model.

Yes, and most people do. Anything inside your band is covered by the flat plan; anything outside it — a closed frontier model called directly, for example — is metered at provider cost with the usual $0 gateway fee.

BYOK is free on every plan, but it is a separate thing: with your own key you pay that provider directly and we add nothing. The unlimited plans cover inference we serve on our own hardware.

There isn't one on tokens. We pass the provider's list price through and itemise it; the gateway line reads $0.00 on every tier. We earn on seats ($10 per user per month for Team), on Enterprise controls and dedicated capacity, on self-host licences, and on our own hosted fleet where we are the provider and take a normal inference margin.

Intelligence and memory. Smart routing (auto:cost, auto:quality, workload-aware directives and fallback chains), org-level analytics broken down per member, key, model and project, 90-day log retention with request inspection and export, a private semantic cache namespace, and enforced spend caps. Tokens stay at provider cost.

Any member with access to the org workspace in a given month. Flat $10 each, no tiering by usage, no per-request platform fee. API keys and service accounts are free — you only pay for humans.

Nothing is retained. Requests and responses are not stored after the response is returned, and analytics are limited to a rolling 24-hour aggregate. If you need history to answer 'what did we spend last month and why did that request go to that model', that is Team.

Yes for reserved capacity. Every tier can call our hosted fleet per token as an ordinary provider. Dedicated endpoints — isolated replicas billed per GPU-hour, with pinned region and precision and a contracted SLA — are an Enterprise entitlement.

Credit balances are float — your money sitting on our balance sheet, often non-refundable. You pay in arrears for tokens you actually consumed, in your own currency, and you can cap spend per key.

Yes. Attach your own provider keys and we charge nothing to route through them on any tier. Several gateways bill a per-request BYOK fee; we do not.

It flows to you. Negotiated rates below list appear as a discount line on your invoice rather than becoming our spread.