Updated November 18, 2026

Cheapest uncensored LLM API.

Every uncensored model you can call by API, sorted by price per million tokens, with how good each one is, so you can see what the cheap ones give up.

Short answer

The cheapest uncensored model worth using is DeepSeek V4.1 Flash Uncensored at $0.12 per million input and $0.45 per million output tokens, ranked #5 (64.3). The cheapest of any kind is Llama 3 8B Lunaris at $0.04 and $0.05, a small model we have not ranked. Best value (score per dollar): GLM 5.3 Flash Uncensored. All on one OpenAI compatible key from OpenUncensored, with no account. The cheapest way to call them: a prepaid per-token key and "routing": {"strategy": "cheapest"}, so each request goes to the cheapest host serving the model, with fallback to the next.

By quality

The cheapest model at each quality bar.

Quality barModel$ / 1M input$ / 1M outputRank (score)
The #1 ranked modelDeepSeek V4.1 Flash 1017 Uncensored$0.33$1.32#1 (69.8)
Cheapest in the top 3GLM 5.3 Flash Uncensored$0.12$0.50#2 (69.7)
Cheapest in the top 10DeepSeek V4.1 Flash Uncensored$0.12$0.45#5 (64.3)
Cheapest ranked modelGLM 4.7 Flash Heretic$0.07$0.40#23 (52.7)
Cheapest model of any kindLlama 3 8B Lunaris$0.04$0.05not ranked
Ranks are the uncensored LLM leaderboard: intelligence, agent tasks, refusals, speed and drift in one score.

Same model

What each host charges.

The same weights cost different amounts on different hosts. Cheapest routing picks the lowest for each request; cached input is what agents mostly pay.

DeepSeek V4.1 Flash UncensoredInputCached inputOutput
Refuseless$0.12$0.025$0.45
UncensuredAI$0.27$0.06$0.95
Venice$0.33$0.008$1.32
NanoGPT$0.33–$1.32
GLM 5.3 Flash UncensoredInputCached inputOutput
Refuseless$0.12$0.025$0.50
heabsy$0.20$0.064$0.80
NanoGPT$0.20–$0.80
UncensuredAI$0.22$0.08$0.80

USD per million tokens, live from the router.

Price list

The 25 cheapest uncensored models.

Sorted by blended price, (3 × input + output) / 4, the cost of a typical request that sends three tokens for every one it gets back.

ModelInputOutputBlendedRankSpeedContext
Llama 3 8B Lunarisl3-lunaris-8b$0.04$0.05$0.042–84 tok/s8K
MythoMax 13Bmythomax-l2-13b$0.08$0.11$0.087––8K
GLM 4.7 Flash Hereticolafangensan-glm-4.7-flash-heretic$0.07$0.40$0.15#2380 tok/s195K
Gemma 4 26B A4B Chimera X Hereticgemma-4-26b-a4b-it-chimeraxheretic$0.12$0.38$0.18–43 tok/s256K
DeepSeek V4.1 Flash Uncensoreddeepseek-v4.1-flash-uncensored$0.12$0.45$0.20#577 tok/s1M
GLM 5.3 Flash Uncensoredglm-5.3-flash-uncensored$0.12$0.50$0.21#2178 tok/s1M
Apex Flash 1 Abliteratedglm-5.3-flash-apex-abliterated$0.15$0.50$0.24#810 tok/s1M
Gemma 4 Uncensoredgemma-4-uncensored$0.16$0.50$0.25#2167 tok/s250K
DeepSeek V4 Pro Qwen3.5 9B Uncensored Hereticdeepseek-v4-pro-qwen3.5-9b-uncensored-heretic$0.28$0.28$0.28–1 tok/s32K
Gemma 4 26B A4B Uncensored TEEgemma-4-26b-a4b-uncensored$0.15$0.70$0.29#2221 tok/s256K
Qwen3.6 35B A3B Uncensored (HauhauCS Aggressive)qwen3.6-35b-a3b-uncensored$0.12$0.80$0.29#1246 tok/s32K
Grayline Qwen3 8Bgrayline-qwen3-8b$0.30$0.30$0.30––32K
Gemma 4 31B Cognitive Unshackledgemma-4-31b-cognitive-unshackled$0.31$0.31$0.31–19 tok/s256K
Gemma 4 31B Gembrain Uncensored Hereticgemma-4-31b-gembrain-uncensored-heretic$0.31$0.31$0.31#1514 tok/s256K
Gemma 4 31B SDFT Heretic RPgemma-4-31b-sdft-heretic-rp$0.31$0.31$0.31–10 tok/s256K
Qwen3.5 27B Blossom V6.4 Derestrictedqwen3.5-27b-blossom-v6.4-derestricted$0.31$0.31$0.31#2021 tok/s256K
Qwen3.5 27B BlueStar v3 Derestrictedqwen3.5-27b-bluestar-v3-derestricted$0.31$0.31$0.31#1924 tok/s256K
Qwen3.5 27B Claude 4.6 Opus Reasoning Distilled Derestrictedqwen3.5-27b-claude-4.6-opus-reasoning-distilled-derestricted$0.31$0.31$0.31#1628 tok/s256K
Qwen3.5 27B Claude 4.6 Opus Reasoning Distilled Derestricted Liteqwen3.5-27b-claude-4.6-opus-reasoning-distilled-derestricted-lite$0.31$0.31$0.31#2526 tok/s256K
Qwen3.5 27B Queen Derestrictedqwen3.5-27b-queen-derestricted$0.31$0.31$0.31#1828 tok/s256K
DeepSeek V4.1 Flash Hereticdeepseek-v4.1-flash-1017-heretic$0.20$0.79$0.35–53 tok/s500K
DeepSeek V4.1 Flash Heretic Thinkingdeepseek-v4.1-flash-1017-heretic-thinking$0.20$0.79$0.35–106 tok/s500K
Cydonia 24B V4.1cydonia-24b-v4.1$0.30$0.50$0.35–53 tok/s128K
Venice Uncensored 1.2venice-uncensored$0.20$0.90$0.38#2613 tok/s125K
UnslopNemo 12Bunslopnemo-12b$0.40$0.40$0.40–33 tok/s1M
USD per million tokens at the cheapest host serving each model. The rest of the catalogue is on the models page.

Pay less

Three ways to cut the bill.

  • Route to the cheapest host. Most models run on several hosts at different prices. "routing": {"strategy": "cheapest"} in the body or X-OU-Strategy: cheapest tries hosts cheapest first (hosts that run the model themselves go before resellers) and falls back to the next one if a host fails.
  • Let the cache work. Prompt tokens a host serves from its cache are billed at its cached price, usually a small fraction of the input price. Agents that resend a long history pay mostly that rate.
  • Pick by score per dollar, not price alone. A cheap model that loops in agent tasks or still refuses costs you retries. The rank column shows what you trade away.
from openai import OpenAI

client = OpenAI(base_url="https://api.openuncensored.com/v1", api_key="ou-...")

r = client.chat.completions.create(
    model="deepseek-v4.1-flash-1017-uncensored",
    messages=[{"role": "user", "content": "Hello"}],
    extra_body={"routing": {"strategy": "cheapest"}},  # cheapest host first, then the next cheapest
)
print(r.choices[0].message.content, r.usage)

Compared with

Renting a GPU.

A 27B model needs one large GPU at roughly $1 to $4 an hour: about $700 to $3,000 a month if it is always on, before your time running it. At the prices above that buys hundreds of millions of tokens, so per token APIs are cheaper for almost everyone below very high, steady volume. Credits are prepaid from $5, never expire, and unused credit is refundable.

Want the best model rather than the cheapest? See the best uncensored LLM API.

FAQ

Questions people ask.

What's the cheapest way to call uncensored open-weight models by API?

A prepaid, per-token API with cheapest-first routing. On OpenUncensored, add "routing": {"strategy": "cheapest"} to a request and it goes to the cheapest host serving that model, falling back to the next if one fails. The cheapest model worth using is DeepSeek V4.1 Flash Uncensored at $0.12 per million input and $0.45 per million output tokens, ranked #5 (64.3). Prompt tokens a host serves from its cache are billed at its cached rate, and there is no subscription or minimum beyond a $5 prepaid key. Renting a GPU only pays off at very high, steady volume.

What is the cheapest uncensored LLM API?

On OpenUncensored the cheapest uncensored model worth using is DeepSeek V4.1 Flash Uncensored at $0.12 per million input and $0.45 per million output tokens, ranked #5 (64.3). The cheapest model of any kind is Llama 3 8B Lunaris at $0.04 and $0.05.

How do I always get the cheapest provider?

Send "routing": {"strategy": "cheapest"} in the request body, or the header X-OU-Strategy: cheapest. Hosts are then tried cheapest first (hosts that run the model themselves go before resellers), falling back to the next one if a host fails.

Are cached prompt tokens cheaper?

Yes. Prompt tokens a provider serves from its cache are billed at that provider's cached price, usually a small fraction of the normal input price. Agents that resend a long history pay mostly the cached rate.

Is it cheaper than renting a GPU?

For most use, yes. A 27B model needs one large GPU at roughly $1 to $4 an hour, about $700 to $3,000 a month if it is always on. At the per token prices on this page that buys hundreds of millions of tokens.

Updated November 18, 2026. Markdown version: /cheapest-uncensored-llm-api.md