Unfil.ai is a pay-as-you-go uncensored API pitched at businesses. Its catalog mixes community abliterated models, Dolphin and Hermes fine-tunes, and its own auto-routing “Unfil Uncensored” tiers. We list the 12 uncensored models it names. There’s no chat app; it’s API only, and you don’t need an email to use it.
This review is based on Unfil’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 Unfil.ai provider page.
The short version
| Best for | API access to abliterated GLM and Llama models without handing over an email |
| Uncensored models | 12 (6 abliterated, 5 fine-tuned, 1 undisclosed) |
| Per-token price | $0.30 / $0.40 to $1.20 / $1.20 per 1M tokens (input / output) |
| Plans | Pay as you go only |
| Privacy score | 50 / 100, 6th of 37 |
| Watch out for | No named company, template text in its legal pages, and a catalog that looks resold |
Models
The 12 models split into three groups.
Abliterated (6):
- Three GLM-4.5-Air Derestricted variants (Iceblink, Iceblink v2, Steam) at $0.50 / $0.50, each with a 131K context window.
- GLM-4.6 Derestricted (355B, MIT license) at $0.55 / $2.10, also 131K.
- Llama 3.3 70B Instruct abliterated (huihui-ai) at $1.10 / $1.20, with only a 16K window.
- DeepSeek R1 Distill Qwen 32B abliterated (huihui-ai) at $0.90 / $1.10, 16K.
Fine-tuned (5):
- Hermes 3 405B, the Nous Research fine-tune of Llama 3.1 405B, at $1.20 / $1.20 with 128K context. It’s the priciest model here.
- Dolphin 2.9.2 Qwen2 72B at $0.60 / $0.70, with an 8K window.
- Gemma 3 27B Nidum Uncensored at $0.60 / $0.70.
- Omega Directive 24B Unslop v2.0, a Mistral Small 3.2 fine-tune, at $0.80 / $0.80.
- Aion-RP Llama 3.1 8B at $0.30 / $0.40, the cheapest.
Undisclosed (1): venice-uncensored:web, a web-search variant at $0.70 / $0.70 with 80K context. The name matches Venice’s Dolphin-based model, but Unfil doesn’t say how it’s served.
The catalog looks dated: the newest models are from the GLM-4.6 era. Unfil also sells mainstream models, which we don’t list. Search the uncensored ones on the Models page.
Pricing
Billing is per token with no upfront cost. For a typical short chat (1,000 tokens in, 500 out):
- about $0.50 per 1,000 chats on Aion-RP 8B
- about $0.75 per 1,000 chats on any of the GLM-4.5-Air Derestricted models
- about $1.80 per 1,000 chats on Hermes 3 405B
Unfil also sells auto-routing Unfil Uncensored tiers, XS to XL, priced in bands from $0.15 to $1.50 per 1M tokens. They don’t say which model handles a given request, so we don’t list them as models.
One mainstream model on the pricing page is listed twice at different prices, with no note on which applies.
Privacy
Unfil scores 50 / 100 on our privacy rubric, which only gives points for documented claims:
| Rubric item | Unfil.ai | Points |
|---|---|---|
| Prompt retention | Says API prompts and responses aren’t stored after processing | 20 / 20 |
| No training on prompts | Stated | 15 / 15 |
| Technical guarantee (TEE / E2EE) | None; policy only | 0 / 20 |
| Signup | None; no email required | 8 / 8 |
| Crypto payments | Accepted | 7 / 7 |
| Jurisdiction | Not stated | 0 / 10 |
| Upstream hosts disclosed | No | 0 / 10 |
| Independent audit | No | 0 / 10 |
That’s 6th in the index. Among services that market themselves as uncensored, only Venice scores higher. Usage logs, including IP addresses, are kept for 30 days, but prompt content isn’t.
The caveats are about who is making these promises. Unfil’s privacy policy and terms still list the website as “localhost:3000”, left over from a template. Neither names a legal entity or jurisdiction. Its model IDs also match NanoGPT’s catalog exactly, which suggests it’s a reseller. We rate that as low to medium confidence. If it’s true, your prompts pass through at least one more provider, and Unfil doesn’t disclose which.
Access and payment
- API only, no chat app. The API docs are marked “coming soon”.
- No email required.
- Card, digital wallets, or crypto.
How it compares
- Cheaper for the same models: Hermes 3 405B is $1 / $1 at OpenRouter (privacy 65) and DeepInfra (58). Aion-RP 8B is $0.20 / $0.20 at Electron Hub (40). Omega Directive 24B is $0.50 / $0.50 at NanoGPT (35).
- Also no signup, more models: NanoGPT needs no account and lists 139 uncensored models.
- Named company, newer models: Venice.ai (55) sells its own Dolphin-based model at $0.20 / $0.90 and GLM 4.7 Flash Heretic at $0.07 / $0.40.
Verdict
On paper, Unfil is one of the better options: a clear no-storage, no-training policy, no email, and crypto accepted. The GLM-4.5-Air Derestricted models at $0.50 / $0.50 with 131K context are reasonably priced.
The rest needs more polish. The legal pages still contain template text, there’s no named company, the API docs aren’t published, and several of its models are cheaper elsewhere. If you value the no-email signup, NanoGPT offers the same with a far larger catalog. If you value the policy, a named company like Venice or OpenRouter gives it more weight.
Compare every Unfil model and price on its provider page, or see every provider in the index.