DeepInfra
US serverless inference platform with an OpenAI-compatible API. Serves Nous Hermes 3 and several classic roleplay fine-tunes alongside mainstream open models, and is the upstream behind many routers.
BILLINGPay as you go per token, funded by card through Stripe. All listed models are served in FP8 except MythoMax (FP16).
| Model | Context | In $/1M | Out $/1M | Per 1K chats |
|---|---|---|---|---|
Sao10K/L3-8B-Lunaris-v1-TurboLlama 3 8B · 8B | 8K | $0.040 | $0.050 | $0.065 |
Gryphe/MythoMax-L2-13bLlama 2 13B · 13B | 4K | $0.40 | $0.40 | $0.60 |
NousResearch/Hermes-3-Llama-3.1-70BLlama 3.1 70B · 70B · Llama 3.1 Community | 131K | $0.70 | $0.70 | $1.05 |
Sao10K/L3.1-70B-Euryale-v2.2Llama 3.1 70B · 70B · Llama 3.1 Community | 131K | $0.85 | $0.85 | $1.27 |
NousResearch/Hermes-3-Llama-3.1-405BLlama 3.1 405B · 405B · Llama 3.1 Community | 131K | $1.00 | $1.00 | $1.50 |
- Method
- Fine-tuned
- Access
- Web chat + OpenAI-compatible API
- Logging
- No logs
- Trains on prompts
- No
- Guarantee
- Policy only
- Jurisdiction
- California, US
- Signup
- Account required
- Payments
- Card
- Data checked
- Oct 11, 2026
PRIVACYInputs are held only in memory and outputs deleted once returned; only metadata is logged. No training on submitted data, except for Google and Anthropic models where their policies apply. Policy ↗
Deep Infra Inc. Not marketed as uncensored; listed because it serves uncensored fine-tunes directly and is the upstream for OpenRouter's Hermes 3 and Euryale. Older Dolphin models remain in its API as deprecated and are left out here.