API
Uncensored AI API pricing: published rates by provider
30 offerings from four providers, grouped by base model. Each rate is the provider's own published figure in USD per 1M tokens, checked 14 November 2026.
Disclosure. Uncensored AI Atlas has a commercial relationship with MolochAI. We may benefit when you sign up or buy credit through our links. MolochAI's rows sit alongside the others, sorted by the same rules, and are marked Partner. How we handle this.
How to compare
Compare rows only when they serve the same thing. A row that names the exact Hugging Face repo can be compared with another row naming the same repo. A provider's own build and a standard model are different weights, so a lower price there is not a like-for-like saving. Context limits also differ: the same variant is listed with 32,768 tokens at one provider and 64,000 at another.
Worked example: the reference workload
The "100K in + 20K out" column prices one request with 100,000 input tokens and 20,000 output tokens:
cost = 100,000 / 1,000,000 × input_rate + 20,000 / 1,000,000 × output_rate e.g. $0.50 and $2.00 per 1M → $0.05 + $0.04 = $0.09
It ignores prompt caching. To price your own mix, including cached tokens and monthly volume, use the token cost calculator or the price comparison tool, which exports CSV.
Qwen3.8-27B
Refusal-reduced offerings built on Qwen3.8-27B. Variants, methods and licenses.
| Provider and model id | What is served | Input / 1M | Output / 1M | Cached input / 1M | Context | Max output | 100K in + 20K out | Checked |
|---|---|---|---|---|---|---|---|---|
| MolochAIPartner qwen3.8-27b-uncensored | Own refusal-reduced build, weights not named Qwen3.8-27B-Uncensored | $0.50 | $2.00 | not listed | 200,704 | 9,000 | $0.0900 | 14 November 2026 |
| Featherless huihui-ai/Huihui-Qwen3.8-27B-abliterated | Names the exact Hugging Face repo | $1.60 | $12.00 | not listed | 32,768 | 32,768 | $0.4000 | 14 November 2026 |
Standard Qwen3.8-27B rates for reference (3)
| Provider and model id | What is served | Input / 1M | Output / 1M | Cached input / 1M | Context | Max output | 100K in + 20K out | Checked |
|---|---|---|---|---|---|---|---|---|
| OpenRouter qwen/qwen3.8-27b | Standard model, not refusal-reduced | $0.15 | $1.875 | $0.0375 | 1,024,000 | Not stated | $0.0525 | 14 November 2026 |
| Featherless Qwen/Qwen3.8-27B | Standard model, not refusal-reduced | $0.40 | $3.00 | not listed | 262,144 | 32,768 | $0.1000 | 14 November 2026 |
| Venice qwen-3-8-27b | Standard model, not refusal-reduced | $0.45 | $3.20 | not listed | 262,144 | 65,536 | $0.1090 | 14 November 2026 |
Gemma 4 26B A4B
Refusal-reduced offerings built on Gemma 4 26B A4B (instruction-tuned). Variants, methods and licenses.
| Provider and model id | What is served | Input / 1M | Output / 1M | Cached input / 1M | Context | Max output | 100K in + 20K out | Checked |
|---|---|---|---|---|---|---|---|---|
| Venice gemma-4-uncensored | Names the exact Hugging Face repo | $0.1625 | $0.50 | not listed | 256,000 | 8,192 | $0.0263 | 14 November 2026 |
| Venice e2ee-gemma-4-26b-a4b-uncensored-p | Names the exact Hugging Face repo | $0.19 | $0.88 | not listed | 64,000 | 4,096 | $0.0366 | 14 November 2026 |
| Featherless coder3101/gemma-4-26B-A4B-it-heretic | Names the exact Hugging Face repo | $0.40 | $0.80 | not listed | 32,768 | 32,768 | $0.0560 | 14 November 2026 |
| Featherless llmfan46/gemma-4-26B-A4B-it-ultra-uncensored-heretic | Names the exact Hugging Face repo | $0.40 | $0.80 | not listed | 32,768 | 32,768 | $0.0560 | 14 November 2026 |
| MolochAIPartner gemma4-26b-a4b-qat | Own refusal-reduced build, weights not named Gemma4-26B-A4B-QAT (refusal-reduced build) | $0.50 | $2.00 | not listed | 200,704 | 9,000 | $0.0900 | 14 November 2026 |
Standard Gemma 4 26B A4B rates for reference (3)
| Provider and model id | What is served | Input / 1M | Output / 1M | Cached input / 1M | Context | Max output | 100K in + 20K out | Checked |
|---|---|---|---|---|---|---|---|---|
| OpenRouter google/gemma-4-26b-a4b-it | Standard model, not refusal-reduced | $0.0675 | $0.225 | $0.0375 | 262,144 | Not stated | $0.0113 | 14 November 2026 |
| Featherless google/gemma-4-26B-A4B-it | Standard model, not refusal-reduced | $0.07 | $0.34 | not listed | 32,768 | 32,768 | $0.0138 | 14 November 2026 |
| Venice google-gemma-4-26b-a4b-it | Standard model, not refusal-reduced | $0.13 | $0.40 | $0.05 | 256,000 | 16,384 | $0.0210 | 14 November 2026 |
GLM-4.7-Flash
Refusal-reduced offerings built on GLM-4.7-Flash. Variants, methods and licenses.
| Provider and model id | What is served | Input / 1M | Output / 1M | Cached input / 1M | Context | Max output | 100K in + 20K out | Checked |
|---|---|---|---|---|---|---|---|---|
| Venice olafangensan-glm-4.7-flash-heretic | Names the exact Hugging Face repo | $0.07 | $0.40 | $0.035 | 200,000 | 24,000 | $0.0150 | 14 November 2026 |
| MolochAIPartner glm-4.7-flash-uncensored | Own refusal-reduced build, weights not named GLM-4.7-Flash-Uncensored | $0.50 | $2.00 | not listed | 200,704 | 9,000 | $0.0900 | 14 November 2026 |
Standard GLM-4.7-Flash rates for reference (2)
| Provider and model id | What is served | Input / 1M | Output / 1M | Cached input / 1M | Context | Max output | 100K in + 20K out | Checked |
|---|---|---|---|---|---|---|---|---|
| OpenRouter z-ai/glm-4.7-flash | Standard model, not refusal-reduced | $0.0605 | $0.40 | not listed | 200,000 | Not stated | $0.0140 | 14 November 2026 |
| Featherless zai-org/GLM-4.7-Flash | Standard model, not refusal-reduced | $0.06525 | $0.40 | not listed | 202,752 | Not stated | $0.0145 | 14 November 2026 |
Llama 3.3 70B
Refusal-reduced offerings built on Llama 3.3 70B Instruct. Variants, methods and licenses.
| Provider and model id | What is served | Input / 1M | Output / 1M | Cached input / 1M | Context | Max output | 100K in + 20K out | Checked |
|---|---|---|---|---|---|---|---|---|
| Featherless huihui-ai/Llama-3.3-70B-Instruct-abliterated | Names the exact Hugging Face repo | $0.65 | $0.75 | not listed | 32,768 | Not stated | $0.0800 | 14 November 2026 |
| MolochAIPartner llama-3.3-70b-instruct-abliterated | Own refusal-reduced build, weights not named Llama-3.3-70B-Instruct-abliterated | $0.50 | $2.00 | not listed | 200,704 | 9,000 | $0.0900 | 14 November 2026 |
Standard Llama 3.3 70B rates for reference (2)
| Provider and model id | What is served | Input / 1M | Output / 1M | Cached input / 1M | Context | Max output | 100K in + 20K out | Checked |
|---|---|---|---|---|---|---|---|---|
| Featherless meta-llama/Llama-3.3-70B-Instruct | Standard model, not refusal-reduced | $0.65 | $0.75 | not listed | 32,768 | Not stated | $0.0800 | 14 November 2026 |
| OpenRouter meta-llama/llama-3.3-70b-instruct | Standard model, not refusal-reduced | $0.71 | $0.71 | not listed | 131,072 | Not stated | $0.0852 | 14 November 2026 |
GLM-5.3-Flash
Refusal-reduced offerings built on GLM-5.3-Flash. Variants, methods and licenses.
| Provider and model id | What is served | Input / 1M | Output / 1M | Cached input / 1M | Context | Max output | 100K in + 20K out | Checked |
|---|---|---|---|---|---|---|---|---|
| MolochAIPartner glm-5.3-flash | Own refusal-reduced build, weights not named GLM-5.3-Flash (refusal-reduced build) | $0.50 | $2.00 | not listed | 200,704 | 9,000 | $0.0900 | 14 November 2026 |
Standard GLM-5.3-Flash rates for reference (2)
| Provider and model id | What is served | Input / 1M | Output / 1M | Cached input / 1M | Context | Max output | 100K in + 20K out | Checked |
|---|---|---|---|---|---|---|---|---|
| OpenRouter z-ai/glm-5.3-flash | Standard model, not refusal-reduced | $0.0399 | $0.0784 | $0.03 | 1,048,576 | Not stated | $0.00556 | 14 November 2026 |
| Venice z-ai-glm-5-3-flash | Standard model, not refusal-reduced | $0.15 | $0.50 | $0.03 | 1,048,576 | 131,072 | $0.0250 | 14 November 2026 |
Other refusal-reduced models
Variants of bases that do not have an Atlas profile yet.
| Provider and model id | What is served | Input / 1M | Output / 1M | Cached input / 1M | Context | Max output | 100K in + 20K out | Checked |
|---|---|---|---|---|---|---|---|---|
| Featherless mlabonne/Meta-Llama-3.1-8B-Instruct-abliterated | Names the exact Hugging Face repo | $0.05 | $0.08 | not listed | 32,768 | Not stated | $0.0066 | 14 November 2026 |
| OpenRouter cognitivecomputations/dolphin-mistral-24b-venice-edition | Names the exact Hugging Face repo | $0.20 | $0.90 | not listed | 128,000 | Not stated | $0.0380 | 14 November 2026 |
| Venice venice-uncensored-1-2 | Names the exact Hugging Face repo | $0.20 | $0.90 | not listed | 128,000 | 8,192 | $0.0380 | 14 November 2026 |
| Featherless p-e-w/Qwen3-4B-Instruct-2507-heretic | Names the exact Hugging Face repo | $0.40 | $0.80 | not listed | 32,768 | Not stated | $0.0560 | 14 November 2026 |
| Featherless mlabonne/gemma-3-27b-it-abliterated | Names the exact Hugging Face repo | $0.40 | $1.20 | not listed | 32,768 | Not stated | $0.0640 | 14 November 2026 |
| Featherless dphn/Dolphin-Mistral-24B-Venice-Edition | Names the exact Hugging Face repo | $0.70 | $1.16 | not listed | 32,768 | Not stated | $0.0932 | 14 November 2026 |
What these prices leave out
- Plans and minimums. Featherless bills per token on its Developer plan, which starts at $50 a month in credits. MolochAI and OpenRouter use prepaid credit. Monthly plans with quotas are not in this table.
- Routing. OpenRouter shows one rate per model but routes to many upstream hosts at different prices; the range is in each row's notes on the provider page.
- Reasoning tokens. Models with a thinking mode bill their reasoning as output tokens on most providers, which can dominate cost.
- Reliability and speed. Not measured. Atlas has not tested uptime, latency or throughput for any provider.
Rates change. Each row links to its source; if a rate is out of date, tell us on the corrections page.