NanoGPT is a pay-per-use aggregator with 640+ models behind one chat app and API. It has the widest selection of abliterated, Heretic and Derestricted checkpoints we found, and we list 139 uncensored models from it: more than any other provider in our index. You don’t need an account, and you can top up with crypto from $0.10.
This review is based on NanoGPT’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 NanoGPT provider page.
The short version
| Best for | Trying many uncensored models from one balance, with no account |
| Uncensored models | 139 (118 fine-tuned, 12 abliterated, 9 undisclosed) |
| Per-token price | $0.03 / $0.10 to $3 / $6 per 1M tokens (input / output) |
| Plans | Pay-as-you-go, optional PRO $12/mo |
| Privacy score | 35 / 100, joint 19th of 37 |
| Watch out for | Upstream hosts’ retention policies still apply, and no company or jurisdiction is named |
Models
NanoGPT is mostly a reseller, so the catalog reads like a map of the uncensored market. Most of it falls into a few groups:
- Community fine-tunes on Gemma 4 and Qwen. Dozens of Gemma 4 31B, Gemma 4 26B A4B and Qwen 3.8 27B roleplay and writing tunes, mostly between $0.10 / $0.30 and $0.306 / $0.306. Several of them carry “(ArliAI)” in the name and match Arli AI’s lineup, including GLM 4.6 Derestricted v5 ($0.40 / $1.50).
- Abliterated models. Classic community checkpoints such as Llama 3.3 70B Instruct abliterated ($0.70 / $0.70) and two DeepSeek R1 distills, plus newer ones like MiMo V2.6 Flash Abliterated ($0.50 / $1.50) with a 1M context window.
- Resold originals. All three Abliteration.ai models at the same prices as the source ($1 / $3 and $3 / $5), Venice Uncensored at $0.40 / $1.80, and five AionLabs models from $0.70 / $1.40.
- Older roleplay favorites. MythoMax 13B, Euryale, Anubis, Magnum, Cydonia and Rocinante, mostly around $0.10 to $0.50 per 1M tokens.
- GLM 5.3 variants. GLM 5.3 Flash Uncensored ($0.20 / $0.80) and GLM 5.3 Uncensored ($1.25 / $2.25), both with 1M context.
One model, Gemma 4 26B A4B Uncensored TEE ($0.15 / $0.70), is labelled as running in a trusted execution environment. It’s a single model, so it doesn’t change NanoGPT’s privacy score.
Nine models don’t say how they were uncensored, including the Manta and Synth series and CyberHeabsy. We mark those “Unverified” until we test them. Search the full list on the Models page.
Pricing
Billing is per token from a prepaid balance, at list price with no deposit fee. A typical short chat (1,000 tokens in, 500 out) costs:
- about $0.08 per 1,000 chats on Synth 2.5 Flash Preview ($0.03 / $0.10), the cheapest model, though its method is undisclosed
- about $0.18 per 1,000 chats on Gemma 4 12B StationKeeper ($0.05 / $0.25)
- about $0.33 per 1,000 chats on the $0.10 / $0.45 Gemma 4 31B fine-tunes
- about $0.90 per 1,000 chats on Hermes 4 Large, the 405B Llama 3.1 fine-tune ($0.30 / $1.20)
The most expensive models are AionLabs’ Aion 3.0 and 3.5 at $3 / $6.
The optional PRO plan is $12/mo and includes 60M input tokens a week on open models that are part of the subscription. Models marked “not in subscription” are always billed per token.
Resale isn’t always cheap. Venice Uncensored costs $0.40 / $1.80 here, about double the $0.20 / $0.90 Venice charges directly.
Privacy
NanoGPT scores 35 / 100 on our privacy rubric, which only gives points for documented claims:
| Rubric item | NanoGPT | Points |
|---|---|---|
| Prompt retention | Not logged or retained during normal inference; chats stored in your browser | 20 / 20 |
| No training on prompts | Not stated | 0 / 15 |
| Technical guarantee (TEE / E2EE) | One TEE-labelled model; everything else relies on policy | 0 / 20 |
| Signup | None | 8 / 8 |
| Crypto payments | Accepted | 7 / 7 |
| Jurisdiction | Not stated | 0 / 10 |
| Upstream hosts disclosed | No | 0 / 10 |
| Independent audit | No | 0 / 10 |
The no-account signup and the no-logging statement are real strengths. The gap is everything after NanoGPT. Its privacy page says final retention is controlled by the third-party providers that actually run the models, and it doesn’t say which host serves which model. The page also doesn’t say whether prompts are used for training, and we couldn’t find a named legal entity or jurisdiction.
Access and payment
- Web chat and an API. No account needed.
- Pay-as-you-go from $0.10 in crypto or $1 by card.
- Optional PRO subscription at $12/mo.
- Chat history is stored in your browser, not on NanoGPT’s servers.
How it compares
- Same models, from the source: Abliteration.ai sells the same three models at the same prices, from a named Delaware company that states it never trains on prompts (privacy score 48). Venice sells Venice Uncensored at half NanoGPT’s price.
- Flat-rate instead of per-token: Arli AI lists 131 uncensored models on plans from $10/mo, all run on its own servers.
- Also no account: PayPerQ (45) takes Lightning payments from $0.10 without an account, with a much smaller uncensored list (9 models).
Verdict
NanoGPT is the easiest way to try a lot of uncensored models without commitment. No account, tiny minimum top-ups, crypto, and the largest catalog we’ve indexed. Many of its models are cheap enough that cost barely matters.
The trade-off is that you’re trusting a chain of hosts you can’t see, run by an operator that isn’t named. If you’ve settled on one model, check whether its original provider sells it cheaper or with stronger privacy terms.
Compare every NanoGPT model and price on its provider page, or see every provider in the index.