Pricing — Parasail
Pricing built for growth.
Production inference that won't break your product or your bank.
Serverless
- Per-token access to 2M+ open models. Call an endpoint and go — zero setup.
- pay as you go
- 2M+ open models
- No minimums
- Community support
Early Access
Dedicated serverless
Private, optimized endpoints priced per million tokens.
Per million tokens
Private optimized endpoints
Per-workload tuning
Dedicated support
Dedicated
Reserved GPUs sized to your roadmap, with negotiated SLAs.
Reserved capacity
Reserved GPUs, sized to you
Negotiated latency SLA
Dedicated support
Batch
High-throughput offline jobs at the lowest per-token rate, on spare fleet capacity.
Lowest rate
Lowest per-token rate
Millions of requests per job
Great for evals & embeddings
Per-token model pricing
Pay only for what you use. No minimums, no rate limits.
| Model | Price per 1M tokens | Input | Output | Cache read |
|---|---|---|---|---|
| GLM-5.2 | $1.40 | $4.40 | $0.26 | |
| GLM-5.1 | $1.40 | $4.40 | $0.26 | |
| GLM-5 | $1.00 | $3.20 | $0.20 | |
| MiniMax M3 | $0.30 | $1.20 | $0.06 | |
| MiniMax M2.5 | $0.30 | $1.20 | $0.03 | |
| DeepSeek V4 Pro | $1.74 | $3.48 | $0.10 | |
| DeepSeek V4 Flash | $0.14 | $0.28 | $0.07 | |
| Kimi K2.7 Code | $0.75 | $3.50 | $0.16 | |
| Kimi K2.6 | $0.75 | $3.50 | $0.16 | |
| Qwen3.6 35B-A3B | $0.15 | $1.00 | $0.05 | |
| Qwen3.5 397B-A17B | $0.50 | $3.60 | $0.30 | |
| Qwen3.5 35B-A3B | $0.15 | $1.00 | $0.05 | |
| Qwen3-Coder-Next | $0.12 | $0.80 | $0.07 | |
| Qwen3-VL 235B-A22B | $0.21 | $1.90 | $0.10 | |
| Qwen3-VL 8B | $0.25 | $0.75 | $0.12 | |
| Qwen3-Next 80B | $0.10 | $1.10 | $0.07 | |
| Qwen3 235B-A22B (2507) | $0.14 | $0.80 | $0.05 | |
| Qwen2.5-VL 72B | $0.80 | $1.00 | $0.40 | |
| Mistral Small 3.2 24B | $0.09 | $0.30 | $0.05 | |
| Llama 4 Maverick (FP8) | $0.35 | $1.00 | $0.17 | |
| Llama 3.3 70B (FP8) | $0.22 | $0.50 | $0.11 | |
| Nemotron 3 Ultra 550B (NVFP4) | $0.50 | $2.50 | $0.10 | |
| MiMo v2.5 | $0.14 | $0.28 | $0.05 | |
| Resemble TTS (English) | $18.50 | — | — | |
| Trinity Large (Thinking) | $0.22 | $0.85 | $0.06 | |
| Gemma 4 26B-A4B | $0.13 | $0.40 | $0.05 | |
| Gemma 4 31B | $0.15 | $0.40 | $0.06 | |
| Skyfall 31B v4.2 | $0.55 | $0.80 | $0.25 | |
| Cydonia 24B v4.1 | $0.30 | $0.50 | $0.15 | |
| gpt-oss-120b | $0.10 | $0.75 | $0.055 | |
| gpt-oss-120b (Fast) | $0.15 | $0.60 | — | |
| gpt-oss-20b | $0.04 | $0.20 | $0.02 | |
| UI-TARS 1.5 7B | $0.10 | $0.20 | $0.10 | |
| Gemma 3 27B | $0.08 | $0.45 | $0.04 | |
| Skyfall 36B v2 (FP8) | $0.55 | $0.80 | $0.25 | |
| BGE-M3 | $0.01 | — | — |
Reserved GPU pricing
Hourly price per GPU, billed by the minute. Volume discounts available.
| Hardware | Memory |
|---|---|
| NVIDIA B300 | 288 GiB |
| NVIDIA B200 | 192 GiB |
| NVIDIA H200 | 141 GiB |
| NVIDIA H100 SXM | 80 GiB |
| NVIDIA RTX PRO 6000 | 96 GiB |
| NVIDIA RTX 5090 | 32 GiB |
Hourly rates available on request · volume discounts available
Self-service batch pricing
Best pricing and speed for your largest jobs. We find a unique fleet configuration, including spot instances, for the best value — a discount on serverless rates, plus an additional discount for cached tokens.
| Model size | Price per 1M tokens | Input | Output | Cache read |
|---|---|---|---|---|
| 0 – 4.1B params | $0.02 | $0.04 | $0.01 | |
| 4.1B – 8.1B params | $0.03 | $0.06 | $0.01 | |
| 8.1B – 16.1B params | $0.04 | $0.10 | $0.02 | |
| 16.1B – 21.1B params | $0.05 | $0.15 | $0.02 | |
| 21.1B – 41.1B params | $0.07 | $0.22 | $0.03 | |
| 41.1B – 80.1B params | $0.13 | $0.35 | $0.06 | |
| 80.1B – 150.1B params | $0.17 | $0.47 | $0.09 | |
| 150.1B – 250.1B params | $0.22 | $0.60 | $0.11 | |
| 250.1B – 500B params | $0.05 | $0.25 | $0.20 | |
| 500B+ params | $0.52 | $1.35 | $0.26 |
Models with special batch pricing
| Model | Input | Output | Cache read |
|---|---|---|---|
| Qwen/Qwen3-235B-A22B-Instruct-2507-FP8 | $0.40 | $0.40 | — |
| Qwen/Qwen3.5-397B-A17B-FP8 | $0.25 | $1.80 | — |
| deepseek-ai/DeepSeek-R1-0528 | $2.00 | $2.00 | — |
| google/gemma-4-31B-it | $0.07 | $0.20 | — |
Start building today
Instantly run any open model — popular or specialized.