The cheapest paid uncensored model you can call by API is Sao10K’s Llama 3 8B Lunaris, a small roleplay fine-tune, at $0.04 / $0.05 per 1M tokens (input / output) on DeepInfra and OpenRouter. That’s about $0.065 per 1,000 short chats, as of October 2026. The cheapest abliterated model you can use pay-as-you-go is GLM 4.7 Flash Heretic on Venice.ai, at about $0.27 per 1,000 chats. A 70B-class model starts at about $0.45, and a handful of models are free.
A “chat” here is 1,000 tokens in and 500 out, so cost per 1,000 chats = input price + half the output price.
Cheapest by category
The cheapest models are all small, and small models do less. We split them into three groups, since a $0.065 8B roleplay model and a $1 70B model aren’t doing the same job.
| Category | Model | Host | Price per 1M (in / out) | Per 1,000 chats |
|---|---|---|---|---|
| Small roleplay fine-tune | Llama 3 8B Lunaris | DeepInfra, OpenRouter | $0.04 / $0.05 | $0.065 |
| Small roleplay fine-tune | Stheno v3.2 8B | Novita, Hugging Face | $0.05 / $0.05 | $0.075 |
| Small roleplay fine-tune | MythoMax 13B | OpenRouter | $0.08 / $0.11 | $0.135 |
| Abliterated | Llama 3.1 8B abliterated | Featherless | $0.05 / $0.08 | $0.09 |
| Abliterated | GLM 4.7 Flash Heretic | Venice | $0.07 / $0.40 | $0.27 |
| Abliterated | Gemma 4 31B abliterated v2 | Featherless | $0.12 / $0.36 | $0.30 |
| Abliterated, 70B | Llama 3.3 70B Instruct abliterated | Featherless | $0.65 / $0.75 | $1.03 |
| Large (70B+) | Hermes 4 70B | AI/ML API | $0.18 / $0.55 | $0.45 |
| Large (70B+) | Anubis 70B | NanoGPT | $0.31 / $0.31 | $0.47 |
| Large (70B+) | Hermes 4 405B (“Hermes 4 Large”) | NanoGPT | $0.30 / $1.20 | $0.90 |
| Large (70B+) | Llama 3.3 Euryale 70B | OpenRouter, NextBit | $0.65 / $0.75 | $1.03 |
A few things the table doesn’t show:
- Some models cost nothing per token. MegaNova serves five free models, including two 70B roleplay fine-tunes, within 550 messages a day on its free tier. Mancer’s MythoLite is free with a 2,560-token context.
- Featherless’s per-token prices need its Developer plan, which starts at $50 a month of credit. Without that commitment, the cheapest abliterated options are GLM 4.7 Flash Heretic on Venice and the same model on PayPerQ at $0.074 / $0.42.
- 8B models have short context. Lunaris and Stheno have an 8,192-token window. GLM 4.7 Flash Heretic has 200,000.
- NanoGPT has lower input prices on Synth 2.5 Flash Preview ($0.03 / $0.10) and Manta Mini ($0.02 / $0.16), but doesn’t say what they’re built on or how they were uncensored.
- Electron Hub lists Hermes 4 70B at $0.10 / $0.40, about $0.30 per 1,000 chats, but Hermes 4 needs one of its paid plans, and we’ve flagged its Hermes 4 405B price as far below every primary host’s.
- Hermes 4 405B costs $1 / $3 at Nebius, the host OpenRouter routes it to, and the same through OpenRouter: about $2.50 per 1,000 chats.
How much quality you give up for the low price is the open question. We haven’t measured it yet; benchmarks are in preparation. Published research on what refusal removal costs is summarized in the abliteration tax.
The same model can cost very different amounts
Resellers and gateways sell many of the same models. Checking a few hosts before you commit can halve your bill.
| Model | Cheapest host | Most expensive host |
|---|---|---|
| Venice Uncensored (Dolphin 24B) | Venice or OpenRouter: $0.65 per 1,000 chats | DemonRoute: $5.12 |
| Gemma 4 31B abliterated v2 | Featherless: $0.30 | DemonRoute: $1.20 |
| Magnum v4 72B | NanoGPT: $3.50 | Mancer: $7.98 |
| Llama 3.3 70B Instruct abliterated | Featherless: $1.03 | DemonRoute: $4.10 |
Gateway markups to know about:
- OpenRouter passes through upstream prices with no inference markup, but buying credits costs 5.5% by card or 5% in crypto.
- Requesty adds 5% to upstream rates.
- PayPerQ charges about 5.5% over upstream.
- AI/ML API prices are almost always exactly 1.3754 times OpenRouter’s, which looks like an undisclosed markup of about 37.5%.
When per-token stops being cheapest
If you chat all day, flat-rate plans can beat even the cheapest per-token price. Arli AI is $10 a month for unlimited use of models up to 31B at 16K context. Infermatic is $9 a month for curated models up to 72B. At $0.45 per 1,000 chats, $9 buys about 20,000 short chats, so per-token stays cheaper until you pass roughly that many a month. Long roleplay sessions use far more tokens per message and reach that point much sooner.
Search any model on the Models page to see every host and price side by side.