Live pricing on OneInfer
Pulled from the OneInfer catalog at request time — rates can change. Capture the timestamp with any production decision.
| Provider | Input $/1M tokens | Output $/1M tokens | Cache read $/1M tokens |
|---|---|---|---|
| novita | $2.00 | $6.00 | $0.250 |
| together_ai | $2.00 | $6.00 | $0.250 |
How to read this table
Each row is labeled by provenance: independently verified listings versus a provider's own vendor-reported number. Routing, quantization, and uptime can all change — record the provider and date alongside any benchmark or latency claim you rely on.
Provider matrix
| Provider | Input $/1M | Output $/1M | Cache read | Batch | Provenance |
|---|---|---|---|---|---|
| Alibaba Qwen (direct) | $2.00 | $6.00 | n/a | $1.00 / $3.00 | vendor_reported |
| DeepInfra | $2.00 | $6.00 | $0.20 | n/a | verified |
| Novita | $2.00 | $6.00 | $0.25 | n/a | verified |
| Vercel AI Gateway | $2.00 | $6.00 | $0.25 | n/a | verified |
| Hugging Face | $2.00 | $6.00 | n/a | n/a | verified |
| Together AI | $2.50 | $6.25 | $0.50 | n/a | verified |
Ready to test the workflow?
Fireworks supports serverless inference
Fireworks lists Qwen3.8-Max for serverless inference and notes fine-tuning is not supported on that listing. A separate, captured rate for Qwen3.8-2.4T-A95B on Fireworks was not in scope at this page's verification — check the live listing before quoting per-token cost.
Routing notes
Hugging Face's hosted route is backed by Novita in this case. Vercel AI Gateway exposes the model under the OpenAI-compatible path used by most existing SDKs. Each provider may run at a different precision (FP8 vs BF16) and a different concurrency ceiling; verify against the live provider dashboard before serving production traffic.