OpenRouter is an API router that sits in front of 80+ inference providers. You get one key and one OpenAI-compatible endpoint, and OpenRouter forwards each request to a host that actually runs the model. We list 22 uncensored models on it. All of them are fine-tunes: Venice’s Dolphin-based model, Nous Research’s Hermes, and a long tail of roleplay models. There are no abliterated checkpoints.
This review is based on OpenRouter’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 OpenRouter provider page.
The short version
| Best for | One API key for many uncensored fine-tunes, at the upstream host’s price |
| Uncensored models | 22 (20 fine-tuned, 2 undisclosed) |
| Per-token price | $0.04 / $0.05 to $3 / $6 per 1M tokens (input / output) |
| Plans | Prepaid credits, no subscription |
| Privacy score | 65 / 100, 1st of 37 |
| Watch out for | Your prompt still reaches the upstream host, under that host’s own policy |
Models
The headline model is Venice: Uncensored, the Dolphin fine-tune of Mistral Small 24B. It costs $0.20 / $0.90 with a 128K context window. Venice is the only provider serving it here, so those requests go to Venice.
The rest of the catalog:
- Hermes: Hermes 4 405B ($1 / $3), Hermes 3 405B ($1 / $1) and Hermes 3 70B ($0.70 / $0.70), all Nous Research fine-tunes of Llama 3.1.
- Sao10K roleplay models: Llama 3 8B Lunaris ($0.04 / $0.05), the cheapest model in the list, plus two Euryale 70B versions ($0.65 / $0.75 and $0.85 / $0.85).
- TheDrummer: Cydonia 24B V4.1 ($0.30 / $0.50), Skyfall 36B V2, and UnslopNemo 12B, which lists a 1M-token context window.
- Older community favorites: MythoMax 13B, ReMM SLERP 13B, WizardLM-2 8x22B and Magnum v4 72B ($2.50 / $5).
- Six AionLabs roleplay models, from the Mini versions ($0.70 / $1.40) up to Aion 3.0 and Aion 3.5 ($3 / $6), the priciest entries.
Two models don’t say how they were made: Mancer’s Weaver (alpha) and MiniMax M2-her. We mark those “Unverified”. As of October 11 OpenRouter doesn’t list Anubis, Rocinante, Lumimaid, Mag Mell, or any abliterated or Heretic model. If that’s what you want, look elsewhere. Search the full list on the Models page.
Pricing
OpenRouter passes through each host’s per-token price with no inference markup. The cost is at the top-up: buying credits by card carries a 5.5% fee (minimum $0.80), and crypto (USDC) carries 5%. In practice you’re paying about 5% over the upstream rate.
A typical short chat (1,000 tokens in, 500 out) costs:
- about $0.07 per 1,000 chats on Llama 3 8B Lunaris
- about $0.65 per 1,000 chats on Venice: Uncensored, the same as buying from Venice directly
- about $2.50 per 1,000 chats on Hermes 4 405B
That pass-through pricing is the useful yardstick for every other gateway in this index. When a reseller lists the same model, compare it with OpenRouter’s number first.
Privacy
OpenRouter scores 65 / 100 on our privacy rubric, the highest in the index:
| Rubric item | OpenRouter | Points |
|---|---|---|
| Prompt retention | Not logged by default | 20 / 20 |
| No training on prompts | Stated | 15 / 15 |
| Technical guarantee (TEE / E2EE) | None; policy only | 0 / 20 |
| Signup | Account required | 3 / 8 |
| Crypto payments | Accepted (USDC) | 7 / 7 |
| Jurisdiction | New York, US | 10 / 10 |
| Upstream hosts disclosed | Yes, per request | 10 / 10 |
| Independent audit | No | 0 / 10 |
Who sees your prompt? Two parties: OpenRouter, which says it doesn’t log prompts or completions by default, and the upstream host that runs the model. OpenRouter names that host for each request, and the host’s own policy applies. For Venice: Uncensored that host is Venice, which also says it doesn’t store API prompts.
One detail to know: OpenRouter offers a 1% discount if you opt in to logging. Leave it off if privacy is the reason you’re here.
The score reflects the router’s own policy and its transparency. It’s not a guarantee about every host behind it. The rubric gives points for naming upstreams because you can then read their policies yourself.
Access and payment
- OpenAI-compatible API, plus a chat interface. An account is required.
- Prepaid USD credits by card or crypto (USDC via Coinbase).
- The free variant of Venice: Uncensored appears to have been retired, so expect to pay for every model in this review.
How it compares
- More abliterated models: Hugging Face Inference Providers (58) reaches community abliterated and Heretic checkpoints, mostly served by Featherless (25 models we list, from $0.05 / $0.08).
- No account: NanoGPT (35) lists 139 models with no signup, and PayPerQ (45) resells several of the same models at about 5.5% over OpenRouter’s prices.
- Direct from the host: Venice (55) sells its own model at the same $0.20 / $0.90, and DeepInfra (58) serves Lunaris at the same $0.04 / $0.05.
Verdict
OpenRouter is the reference point for this category. It doesn’t mark up inference, it names the host for every request, and it has the best privacy score in our index. The catalog is broad on fine-tunes: Hermes, Euryale, Cydonia and the Aion line are all here.
Its gaps are clear too. There’s no abliterated model at all, there’s no technical guarantee, and you still depend on each upstream host’s policy. If you want abliterated checkpoints or no account, use one of the alternatives above.
Compare every OpenRouter model and price on its provider page, or see every provider in the index.