Parasail
Inference providerparasail.io ↗
Models on Parasail · prices & measurements for models served by Parasail’s APIUSD / MTOK · VERIFIED 23 MINUTES AGO
| MODEL | INPUT | OUTPUT | CONTEXT | QUANT | TTFT | TPS | SMOKE | VS BEST ELSEWHERE |
|---|---|---|---|---|---|---|---|---|
 gpt-oss-20bOpenAI | $0.03 | $0.15 | 131K | fp4 | 0.89s° | 52° tok/s | 5/5 | $0.018 elsewhere |
 Mistral NemoMistral | $0.03 | $0.03 | 131K | fp8 | 0.9s° | 106° tok/s | — | $0.018 elsewhere |
 Llama 3 8B LunarisSao10KBEST PRICE | $0.04 | $0.05 | 8K | bf16 | 0.28s° | 207° tok/s | — | cheapest here |
 Llama 3.2 3B InstructMetaBEST PRICE | $0.05 | $0.33 | 131K | bf16 | 0.69s° | 330° tok/s | — | cheapest here |
 Gemma 3 27BGoogleBEST PRICE | $0.08 | $0.45 | 131K | fp8 | 1.4s° | 71° tok/s | 4/5 | cheapest here |
 MythoMax 13BGrypheBEST PRICE | $0.08 | $0.11 | 4K | fp16 | ✕ | ✕ | — | cheapest here |
| Mistral Small 3.2 24BMistral | $0.09 | $0.30 | 131K | bf16 | 1.5s° | 64° tok/s | — | $0.075 elsewhere |
 Qwen3.5-9BQwen | $0.10 | $0.25 | 262K | bf16 | ✕ | ✕ | — | $0.08 elsewhere |
| Qwen3 Next 80B A3B InstructQwen | $0.10 | $1.10 | 262K | fp8 | ✕ | ✕ | — | $0.09 elsewhere |
| gpt-oss-120bOpenAI | $0.10 | $0.75 | 131K | fp4 | 0.57s° | 236° tok/s | — | $0.03 elsewhere |
 UI-TARS 7B ByteDance | $0.10 | $0.20 | 128K | bf16 | ✕ | ✕ | — | only provider |
| Qwen3 Coder NextQwenBEST PRICE | $0.12 | $0.80 | 262K | bf16 | ✕ | ✕ | — | cheapest here |
| Gemma 4 26B A4B Google | $0.13 | $0.40 | 262K | bf16 | 1.3s° | 30° tok/s | — | $0.042 elsewhere |
 DeepSeek V4 Flash 0731DeepSeek | $0.14 | $0.28 | 1.05M | fp8 | ✕ | ✕ | — | $0.01 elsewhere |
| DeepSeek V4 Flash 0423DeepSeek | $0.14 | $0.28 | 1.05M | fp8 | 1.3s° | 101° tok/s | — | $0.017 elsewhere |
| Qwen3 235B A22B Instruct 2507Qwen | $0.14 | $0.80 | 131K | fp8 | 1.1s° | 54° tok/s | 5/5 | $0.087 elsewhere |
 GLM 5.3 FlashZ.ai | $0.15 | $0.50 | 1.05M | fp4 | ✕ | ✕ | — | $0.032 elsewhere |
| Qwen3.6 35B A3BQwen | $0.15 | $1.00 | 262K | fp8 | ✕ | ✕ | 2/5 | $0.05 elsewhere |
| Gemma 4 31BGoogle | $0.15 | $0.40 | 262K | fp8 | 1.4s° | 38° tok/s | — | $0.09 elsewhere |
| Qwen3.5-35B-A3BQwen | $0.15 | $1.00 | 262K | fp8 | ✕ | ✕ | 3/5 | $0.08 elsewhere |
| Qwen3 VL 235B A22B InstructQwen | $0.21 | $1.90 | 131K | fp8 | 1.3s° | 74° tok/s | — | $0.20 elsewhere |
| Llama 3.3 70B InstructMeta | $0.22 | $0.50 | 131K | fp8 | ✕ | ✕ | 4/5 | $0.10 elsewhere |
| Qwen3.8 27BQwen | $0.24 | $2.20 | 262K | fp8 | 0.96s° | 38° tok/s | 4/5 | $0.024 elsewhere |
| Qwen3 VL 8B InstructQwen | $0.25 | $0.75 | 262K | bf16 | 1.2s° | 70° tok/s | — | $0.117 elsewhere |
| DeepSeek V4.1 FlashDeepSeek | $0.30 | $1.20 | 1.05M | fp8 | 0.82s° | 293° tok/s | — | $0.018 elsewhere |
 MiniMax M3MiniMax | $0.30 | $1.20 | 1.05M | fp8 | 0.67s° | 213° tok/s | 3/5 | $0.23 elsewhere |
 Cydonia 24B V4.1TheDrummer | $0.30 | $0.50 | 131K | bf16 | 2.5s° | 13° tok/s | — | only provider |
| Llama 4 MaverickMeta | $0.35 | $1.00 | 524K | fp8 | 0.65s° | 52° tok/s | 5/5 | $0.188 elsewhere |
| UnslopNemo 12BTheDrummer | $0.40 | $0.40 | 1.02M | bf16 | ✕ | ✕ | — | only provider |
| DeepSeek V4 Pro 0423DeepSeek | $0.45 | $3.48 | 1.05M | fp8 | 1.5s° | 70° tok/s | — | $0.30 elsewhere |
| Qwen3.5 397B A17BQwen | $0.50 | $3.60 | 262K | fp8 | ✕ | ✕ | 3/5 | $0.39 elsewhere |
| Skyfall 36B V2TheDrummer | $0.55 | $0.80 | 33K | fp8 | 1.3s° | 74° tok/s | — | only provider |
 Kimi K2.6Moonshot AI | $0.75 | $3.50 | 262K | int4 | ✕ | ✕ | — | $0.465 elsewhere |
| Qwen2.5 VL 72B InstructQwen | $0.80 | $1.00 | 128K | fp8 | 1.4s° | 46° tok/s | — | only provider |
| DeepSeek V4 Pro 0813DeepSeek | $1.32 | $3.96 | 1.05M | fp8 | 0.88s° | 69° tok/s | — | $0.158 elsewhere |
| GLM 5.3Z.ai | $1.40 | $4.40 | 1.05M | fp8 | 0.97s° | 244° tok/s | — | $0.039 elsewhere |
| GLM 5.2Z.ai | $1.40 | $4.40 | 262K | fp4 | 1s° | 76° tok/s | — | $0.03 elsewhere |
| Kimi K3Moonshot AI | $2.60 | $13.00 | 1.05M | fp4 | 0.91s° | 235° tok/s | — | $0.68 elsewhere |
TTFT/TPS: ModelIndex-measured medians (P50, trailing 21d) for Parasail’s endpoints · ° = measured via OpenRouter routing · ✕ = probed, no measurable stream · SMOKE: of 5 capability probes passed · methodology on each model page