Novita AI is a GPU cloud and serverless LLM API with OpenAI-compatible endpoints. Most of its catalog is mainstream open models, and it doesn’t sell itself as an uncensored service. We list the 4 models that are uncensored: classic Llama-based roleplay fine-tunes, including one Novita itself describes as “the uncensored llama3 model.”
This review is based on Novita’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 Novita provider page.
The short version
| Best for | Very cheap per-token access to small, well-known roleplay models |
| Uncensored models | 4 (all fine-tuned) |
| Per-token price | $0.05 / $0.05 to $1.48 / $1.48 per 1M tokens (input / output) |
| Plans | Pay as you go, $10 minimum top-up |
| Privacy score | 48 / 100, joint 7th of 37 |
| Watch out for | An 8K context limit on every model, and a 70B Euryale that costs about twice what other hosts charge |
Models
Novita’s uncensored selection is small and dated. All four models are community fine-tunes of Llama 2 or Llama 3:
- Stheno v3.2 8B ($0.05 / $0.05): Sao10K’s Llama 3 8B roleplay fine-tune, with an 8K context window. The cheapest model here.
- MythoMax L2 13B ($0.09 / $0.09): Gryphe’s Llama 2 13B roleplay model, with a 4K context window.
- Hermes 2 Pro 8B ($0.14 / $0.14): an earlier Nous Research Hermes, built on Llama 3 8B, with 8K context.
- Euryale v2.1 70B ($1.48 / $1.48): Sao10K’s Llama 3 70B fine-tune, with 8K context. This is the one Novita’s own model description calls “the uncensored llama3 model.”
There are no abliterated models, and nothing built on Llama 3.1 or later, Mistral, Qwen or Gemma. The largest context window is 8K tokens, which is short for long roleplay sessions. Compare these models across hosts on the Models page.
Pricing
Novita is pay as you go per token. You top up by card through Stripe, with a $10 minimum. A typical short chat (1,000 tokens in, 500 out) costs:
- about $0.08 per 1,000 chats on Stheno 8B
- about $0.14 per 1,000 chats on MythoMax
- about $0.21 per 1,000 chats on Hermes 2 Pro 8B
- about $2.22 per 1,000 chats on Euryale v2.1 70B
The small models are among the cheapest in our index. MythoMax at $0.09 costs less than a quarter of DeepInfra’s $0.40.
Euryale goes the other way. Electron Hub lists the same v2.1 at $0.70 / $0.80. NextBit sells the newer Llama 3.3 Euryale at $0.65 / $0.75 with a 128K context window.
Privacy
Novita scores 48 / 100 on our privacy rubric, which only gives points for documented claims:
| Rubric item | Novita | Points |
|---|---|---|
| Prompt retention | “Zero Data Retention”: content not kept beyond what’s needed to generate output | 20 / 20 |
| No training on prompts | Not used for training by default | 15 / 15 |
| Technical guarantee (TEE / E2EE) | None; relies on policy | 0 / 20 |
| Signup | Account required | 3 / 8 |
| Crypto payments | Not accepted | 0 / 7 |
| Jurisdiction | Delaware, US | 10 / 10 |
| Upstream hosts disclosed | Not documented | 0 / 10 |
| Independent audit | No | 0 / 10 |
That’s joint 7th of 37, level with Parasail, Abliteration.ai and AionLabs. The retention terms are clear. Two qualifiers come from Novita’s own terms: automated safety screening of content is allowed, and the no-training commitment applies “by default.”
Access and payment
- A web chat plus an OpenAI-compatible API, with an account required.
- Card only, through Stripe, with a $10 minimum top-up.
- Novita is also a Hugging Face Inference Provider, and Requesty routes to its Stheno and MythoMax at the same prices. Using Novita through a gateway adds that gateway’s policy on top of Novita’s.
How it compares
- Longer MythoMax context: OpenRouter (privacy 65) has MythoMax with 8K context, double Novita’s 4K, at $0.08 / $0.11. It also has Lunaris 8B at $0.04 / $0.05.
- Better 70B value: NextBit (privacy 58) sells Llama 3.3 Euryale 70B at $0.65 / $0.75 with 128K context, from its own hardware in Spain.
- Newer models: Featherless (25 models) and Venice (13 models) carry abliterated and current-generation fine-tunes.
Verdict
Novita is worth knowing about for one thing: cheap small roleplay models. Stheno 8B and MythoMax cost cents per thousand chats, and they come from a named US company with zero-retention terms.
Beyond that, the selection is thin. Every model is a generation or two old, none goes past 8K context, and the 70B option is priced well above other hosts. If you need a bigger or newer model, look elsewhere.
Compare every Novita model and price on its provider page, or see every provider in the index.