Inference provider
Darkbloom: every model it hosts
3 endpoints tracked. Hosts may use different quantizations, context limits, and serving defaults, so compare price together with reliability and configuration. On this host family, the same model costs up to 4× more depending on which endpoint serves it.
Verified 2026-10-21 · source: OpenRouter public API
| Model | Input /1M | Output /1M | Cache read | Context | Uptime 1d | Throughput |
|---|---|---|---|---|---|---|
| NVIDIA: Nemotron 3.5 Lightningint4 | $0.025 | $0.120 | $0.013 | 262K | 99.9% | — |
| Qwen: Qwen3.6 35B A3Bfp4 | $0.050 | $0.700 | $0.025 | 262K | 100.0% | — |
| Xiaomi: MiMo-V2.6-Flashfp4 | $0.100 | $0.280 | $0.050 | 1049K | 97.7% | — |
Prices are list prices at build time; providers run promotions. Uptime/throughput are OpenRouter's rolling measurements. Compare hosts on the providers index.