Not always, and you usually can’t check for yourself. Many uncensored models are resold through other services: Abliteration.ai’s models are sold by Venice and NanoGPT, Venice’s Dolphin fine-tune by OpenRouter, NanoGPT and Unfil, and Arli AI’s GLM-4.6 Derestricted by NanoGPT. Markups run from Requesty’s stated 5% through AI/ML API’s prices of exactly 1.3754 times OpenRouter’s to DemonRoute’s at exactly 4 times Featherless’s. A few services sell renamed or undisclosed models. We haven’t yet verified what any endpoint actually serves.
Prices are $in / $out per 1M tokens (input / output), as of October 2026.
Who resells whom
| Original | Source price | Also sold by |
|---|---|---|
| Venice Uncensored (Dolphin Mistral 24B Venice Edition) | $0.20 / $0.90 at Venice | OpenRouter $0.20 / $0.90 (Venice is the only provider behind it), PayPerQ $0.211 / $0.9495, AI/ML API $0.2751 / $1.2379, NanoGPT $0.40 / $1.80, Unfil $0.70 / $0.70, DemonRoute $2.80 / $4.64 |
| Abliteration.ai’s abliterated-model-large-v2 (GLM-5.3) | $3 / $5 at Abliteration.ai | Venice and NanoGPT, both $3 / $5 |
| Arli AI’s GLM-4.6-Derestricted-v5 | Flat-rate plans at Arli AI | NanoGPT $0.40 / $1.50, Electron Hub $0.60 / $1.20 |
| AionLabs’ Aion 3.5 Mini | $0.70 / $1.40 at AionLabs | OpenRouter and NanoGPT at the same price; Venice $0.875 / $1.75 |
Reselling isn’t a problem in itself. It can be the easiest way to reach a model, and some resellers charge the source price. But every extra hop is another company that sees your prompt, and the price can change a lot along the way. The Dolphin model’s weights are also public, so Featherless serves its own copy at $0.70 / $1.16 (per-token prices need its $50/mo Developer plan).
How big the markups are
| Gateway | What our data shows |
|---|---|
| OpenRouter | No markup on model prices. Buying credits costs 5.5% by card or 5% in crypto |
| Hugging Face | No markup on upstream provider rates |
| Requesty | Upstream rates plus a 5% pay-as-you-go markup |
| PayPerQ | About 5.5% above upstream: exactly 1.055 times Venice’s and OpenRouter’s prices on all 9 models we list |
| NanoGPT | Abliteration.ai and Aion models at source prices, but Venice Uncensored at about double Venice’s |
| AI/ML API | For 9 of the 10 models it shares with OpenRouter, exactly 1.3754 times OpenRouter’s price. It doesn’t name its upstream |
| DemonRoute | Exactly 4 times Featherless’s price on the four checkpoints both list. It doesn’t name its upstream hosts |
A matching multiplier doesn’t prove where a gateway buys. AI/ML API’s prices are consistent with reselling OpenRouter at about 37.5% on top, but it doesn’t say so, and DemonRoute doesn’t say whether it buys from Featherless.
Red flags: prices too low, names that don’t add up
Implausibly cheap. Electron Hub lists Hermes 4 405B at $0.20 / $0.80. Nebius, OpenRouter and Requesty all list it at $1 / $3. Electron Hub doesn’t explain the difference, and we couldn’t match its price to any source. NanoGPT’s Hermes 4 Large (405B) at $0.30 / $1.20 shows the same pattern.
Renamed frontier models. Unjailed AI sells 20 renamed frontier models far below list price. NoLimitGPT sells models named Opus, Sonnet, GPT and Grok. Neither explains where they come from, and we leave them out of the index.
Undisclosed models. NoTrack runs one model whose base isn’t disclosed. Uncensored.chat’s two models don’t name a base, and any model name it doesn’t recognize is served by its v3 model. NoFilterGPT calls its model “based on the GPT architecture” while its terms mention an unnamed upstream vendor. Across the whole index, 61 of 504 models don’t say how they were uncensored; we mark those Unverified.
Look-alike catalogs. Unfil’s model IDs match NanoGPT’s exactly, which suggests Unfil resells NanoGPT. We rate that low to medium confidence.
How to check before you pay
- Search the model name on the Models page. Every host we list for a model shows up side by side, with prices. Try Hermes 405B to see the spread.
- Buy from the source when it’s no more expensive. Venice’s own price for its Dolphin model is the lowest we found; OpenRouter matches it because Venice serves the requests. Abliteration.ai’s models cost the same at the source as at resellers.
- Prefer hosts that name the upstream. OpenRouter shows which provider serves each request. Hugging Face, Requesty and PayPerQ also disclose their upstream hosts.
- Treat an unexplained discount as a question. If a reseller is far below the original host, ask what it’s actually serving.
What we haven’t verified yet
Everything above comes from published prices, model lists and policies. We haven’t yet compared what two hosts actually return for the same model name. That’s what our planned endpoint tests are for: the same weights can behave differently by host, and a service that’s really a mainstream model behind a jailbreak prompt only shows up when you test the endpoint. Our methodology also includes a weekly fidelity check to flag silent model swaps. How do you measure “uncensored”? explains the approach. Benchmarks are in preparation.
Compare every host and price for a model on the Models page.