DeepInfra is a US serverless inference platform with an OpenAI-compatible API. It doesn’t market anything as uncensored. It’s in our index because it serves several uncensored fine-tunes directly, and because it’s the upstream host behind OpenRouter’s Hermes 3 and Euryale. We list 5 models: Nous Research’s Hermes 3 in two sizes, plus three classic roleplay fine-tunes.
This review is based on DeepInfra’s published pricing, model list, and policies, checked on October 11, 2026. Benchmark results from our own runs will be added when they’re complete. Live numbers are on the DeepInfra provider page.
The short version
| Best for | Cheap, direct access to Hermes 3 and a few roleplay classics, from a named company that doesn’t log prompts |
| Uncensored models | 5 (all fine-tuned) |
| Per-token price | $0.04 / $0.05 to $1 / $1 per 1M tokens (input / output) |
| Plans | Pay as you go, no subscription |
| Privacy score | 58 / 100, joint 2nd of 37 |
| Watch out for | A small, older selection, short context on the small models, and card-only payment |
Models
DeepInfra’s uncensored lineup is a handful of community fine-tunes built on Llama. There are no abliterated models, and nothing DeepInfra trained itself.
- Hermes 3 405B ($1 / $1) and Hermes 3 70B ($0.70 / $0.70): Nous Research’s fine-tunes of Llama 3.1, both with 128K context windows. These are the most capable models in the set.
- Euryale v2.2 70B ($0.85 / $0.85): Sao10K’s Llama 3.1 roleplay fine-tune, also with 128K context.
- MythoMax L2 13B ($0.40 / $0.40): Gryphe’s Llama 2 13B roleplay model. Its context window is only 4K tokens.
- Lunaris v1 Turbo 8B ($0.04 / $0.05): Sao10K’s Llama 3 8B fine-tune with an 8K context window, and the cheapest model here.
All of these are served in FP8 except MythoMax, which runs in FP16. Older Dolphin models are still in DeepInfra’s API but marked deprecated, so we leave them out. Search the lineup on the Models page.
Pricing
DeepInfra is pay as you go per token, funded by card through Stripe. There are no plans to choose between. A typical short chat (1,000 tokens in, 500 out) costs:
- about $0.07 per 1,000 chats on Lunaris 8B
- about $0.60 per 1,000 chats on MythoMax
- about $1.05 per 1,000 chats on Hermes 3 70B
- about $1.50 per 1,000 chats on Hermes 3 405B
Hermes 3 405B at $1 / $1 is the lowest direct price we list for that model. Venice charges $1.10 / $3.00 and Unfil $1.20 / $1.20. OpenRouter charges the same $1 / $1 because it routes to DeepInfra.
MythoMax is the exception. At $0.40 / $0.40 it costs more than four times what Novita charges ($0.09 / $0.09) for the same model.
Privacy
DeepInfra scores 58 / 100 on our privacy rubric, which only gives points for documented claims:
| Rubric item | DeepInfra | Points |
|---|---|---|
| Prompt retention | Inputs held only in memory, outputs deleted once returned; metadata only is logged | 20 / 20 |
| No training on prompts | Stated (Google and Anthropic models follow those vendors’ policies) | 15 / 15 |
| Technical guarantee (TEE / E2EE) | None; relies on policy | 0 / 20 |
| Signup | Account required | 3 / 8 |
| Crypto payments | Not accepted | 0 / 7 |
| Jurisdiction | California, US | 10 / 10 |
| Upstream hosts disclosed | Yes | 10 / 10 |
| Independent audit | No | 0 / 10 |
That puts DeepInfra joint 2nd of 37, level with Hugging Face and NextBit. Only OpenRouter (65) scores higher. Its data privacy page is short and specific, which is more than many uncensored-branded services offer.
The training exception for Google and Anthropic models doesn’t affect any model in this review. The gaps are the ones most hosts share: no confidential computing or encryption you can verify, no independent audit, and no way to pay without a card.
Access and payment
- A web chat plus an OpenAI-compatible API, with an account required.
- Card only, through Stripe.
- Run by Deep Infra Inc., based in California.
- These models sit in a general catalog next to mainstream ones. DeepInfra doesn’t promote them as uncensored, so don’t expect an uncensored section or guidance.
How it compares
- Same models, more choice: OpenRouter (privacy 65) sells DeepInfra’s Hermes 3 and Euryale at the same prices, lists 22 uncensored models, and names the upstream host for every request.
- Cheaper small models: Novita has MythoMax at $0.09 / $0.09 and Stheno 8B at $0.05 / $0.05, with a privacy score of 48.
- More uncensored models: Venice (13 models, privacy 55) and Featherless (25 models, privacy 23) carry abliterated and newer fine-tunes that DeepInfra doesn’t.
Verdict
DeepInfra is a solid, low-cost place to run Hermes 3 and a few classic roleplay models. It’s a named US company with one of the clearest no-logging policies in the index. If you already call these models through OpenRouter, you’re probably already using DeepInfra.
Look elsewhere if you want abliterated models, newer fine-tunes, long context on a small model, or crypto payment. And check MythoMax pricing before you commit: other hosts sell it for much less.
Compare every DeepInfra model and price on its provider page, or see every provider in the index.