Updated November 18, 2026

OpenAI-compatible API for abliterated models.

Use abliterated, heretic and uncensored models with the OpenAI SDK and every tool built on it, by changing one base URL.

Short answer

Yes. OpenUncensored is an OpenAI compatible API (https://api.openuncensored.com/v1) for 62 uncensored models, 26 of them abliterated (plain abliteration, heretic, derestricted or orthogonalized), plus 21 image models. Point the OpenAI SDK's base URL at it, use a prepaid key, and keep the rest of your code. The same key also works with the OpenAI Responses API and the Anthropic Messages API. The best abliterated model on our tests is Abliterated Large Flash (abliterated-large-flash), #6 (64.0).

Endpoints

What the API speaks.

APIPathUsed by
OpenAI Chat Completionshttps://api.openuncensored.com/v1/chat/completionsThe OpenAI SDKs, LangChain, LiteLLM, the Vercel AI SDK, Cline, Cursor, OpenCode, most apps
OpenAI Responseshttps://api.openuncensored.com/v1/responsesCodex, the OpenAI Agents SDK
Anthropic Messageshttps://api.openuncensored.com/v1/messagesClaude Code, the Anthropic SDKs
Imageshttps://api.openuncensored.com/v1/images/generations21 uncensored image models
Modelshttps://api.openuncensored.com/v1/modelsEvery model id, free and without a key

Code

Change the base URL.

Buy a prepaid key on the credits page with no account (the key is the account), or point your agent at agents.md and let it buy one.

from openai import OpenAI

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

r = client.chat.completions.create(
    model="abliterated-large-flash",
    messages=[{"role": "user", "content": "Hello"}],
)
print(r.choices[0].message.content)

Models

Abliterated models on the API.

ModelModel idMethodRank (score)AnswersSpeed$ / 1M in · out
Abliterated Large Flashabliterated-large-flashAbliterated#6 (64.0)92%184 tok/s$1.00 · $2.00
Apex Flash 1 Abliteratedglm-5.3-flash-apex-abliteratedAbliterated#8 (63.2)86%10 tok/s$0.15 · $0.50
MiMo V2.6 Flash Abliteratedmimo-v2.6-flash-abliteratedAbliterated#10 (62.6)83%99 tok/s$0.50 · $1.50
Abliterated Large v2abliterated-large-v2Abliterated#11 (62.0)94%283 tok/s$3.00 · $5.00
Kimi K3.1 Pro Uncensoredkimi-k3.1-pro-uncensoredAbliterated#14 (59.7)97%59 tok/s$2.50 · $5.00
Gemma 4 31B Gembrain Uncensored Hereticgemma-4-31b-gembrain-uncensored-hereticHeretic (automated abliteration)#15 (58.5)89%14 tok/s$0.31 · $0.31
Qwen3.5 27B Claude 4.6 Opus Reasoning Distilled Derestrictedqwen3.5-27b-claude-4.6-opus-reasoning-distilled-derestrictedDerestricted#16 (57.8)100%28 tok/s$0.31 · $0.31
Inkling 2 Small Uncensoredinkling-2-small-uncensoredOrthogonalized#17 (57.7)94%395 tok/s$0.60 · $2.40
Qwen3.5 27B Queen Derestrictedqwen3.5-27b-queen-derestrictedDerestricted#18 (57.3)100%28 tok/s$0.31 · $0.31
Qwen3.5 27B BlueStar v3 Derestrictedqwen3.5-27b-bluestar-v3-derestrictedDerestricted#19 (56.4)100%24 tok/s$0.31 · $0.31
Qwen3.5 27B Blossom V6.4 Derestrictedqwen3.5-27b-blossom-v6.4-derestrictedDerestricted#20 (56.3)100%21 tok/s$0.31 · $0.31
GLM 4.7 Flash Hereticolafangensan-glm-4.7-flash-hereticHeretic (automated abliteration)#23 (52.7)100%80 tok/s$0.07 · $0.40
GLM 4.6 Derestricted v5glm-4.6-derestricted-v5Derestricted#24 (50.2)100%31 tok/s$0.40 · $1.50
Qwen3.5 27B Claude 4.6 Opus Reasoning Distilled Derestricted Liteqwen3.5-27b-claude-4.6-opus-reasoning-distilled-derestricted-liteDerestricted#25 (49.8)89%26 tok/s$0.31 · $0.31
Qwen 2.5 32B Abliteratedqwen2.5-32b-instruct-abliteratedAbliterated#28 (44.6)100%19 tok/s$0.70 · $0.70
Llama 3.3 70B Instruct abliteratedllama-3.3-70b-instruct-abliteratedAbliterated#29 (41.7)100%19 tok/s$0.70 · $0.70
DeepSeek R1 Llama 70B Abliterateddeepseek-r1-distill-llama-70b-abliteratedAbliterated#30 (40.4)100%18 tok/s$0.70 · $0.70
DeepSeek R1 Qwen Abliterateddeepseek-r1-distill-qwen-32b-abliteratedAbliterated#31 (40.0)100%20 tok/s$1.40 · $1.40
Neural Daredevil 8B abliteratedneuraldaredevil-8b-abliteratedAbliterated#32 (35.8)100%32 tok/s$0.44 · $0.44
Llama 3 70B abliteratedmeta-llama-3-70b-instruct-abliterated-v3.5Abliterated#33 (34.4)100%17 tok/s$0.70 · $0.70
DeepSeek V4.1 Flash Hereticdeepseek-v4.1-flash-1017-hereticHeretic (automated abliteration)not ranked (72.7)94%53 tok/s$0.20 · $0.79
Abliterated Modelabliterated-modelAbliteratednot ranked (68.4)100%114 tok/s$1.00 · $3.00
Gemma 4 26B A4B Chimera X Hereticgemma-4-26b-a4b-it-chimeraxhereticHeretic (automated abliteration)not ranked (61.9)100%43 tok/s$0.12 · $0.38
Gemma 4 31B SDFT Heretic RPgemma-4-31b-sdft-heretic-rpHeretic (automated abliteration)not ranked (61.3)100%10 tok/s$0.31 · $0.31
DeepSeek V4 Pro Qwen3.5 9B Uncensored Hereticdeepseek-v4-pro-qwen3.5-9b-uncensored-hereticHeretic (automated abliteration)not scored yet–1 tok/s$0.28 · $0.28
DeepSeek V4.1 Flash Heretic Thinkingdeepseek-v4.1-flash-1017-heretic-thinkingHeretic (automated abliteration), reasoningnot scored yet–106 tok/s$0.20 · $0.79
Abliteration removes the refusal direction from a model's weights; heretic searches the edit automatically. Uncensored fine-tunes are on the models page, 62 text models in all.

Details

What else carries over.

  • Streaming in the OpenAI server-sent events format, with a usage chunk at the end.
  • Tool calls through the standard tools field, so agent frameworks work unchanged.
  • Fallback across 8 hosts: a failed request moves to the next host serving the model. Pick cheapest, fastest or default routing with "routing": {"strategy": ...} (docs).
  • Cached prompt pricing: tokens a host serves from its cache are billed at its cached rate.
  • Prepaid keys: when the credit runs out requests stop with 402 instead of running up a bill.

FAQ

Questions people ask.

Is there an OpenAI-compatible API for abliterated models?

Yes. OpenUncensored is an OpenAI compatible API (https://api.openuncensored.com/v1) for 62 uncensored models, 26 of them abliterated (plain abliteration, heretic, derestricted or orthogonalized), plus 21 image models. Point the OpenAI SDK's base URL at it, use a prepaid key, and keep the rest of your code. The same key also works with the OpenAI Responses API and the Anthropic Messages API. The best abliterated model on our tests is Abliterated Large Flash (abliterated-large-flash), #6 (64.0).

Which abliterated model is best?

Abliterated Large Flash (abliterated-large-flash) ranks highest of the abliterated models, #6 (64.0). Across every uncensored model, fine-tunes and abliterations included, the leader is DeepSeek V4.1 Flash 1017 Uncensored; see the leaderboard.

Does it work with LangChain, LiteLLM and the Vercel AI SDK?

Yes. Anything that takes an OpenAI base URL works: set it to https://api.openuncensored.com/v1 and the API key to your OpenUncensored key.

Do streaming and tool calls work?

Yes. Streaming is server-sent events in the OpenAI format, and tool calls use the standard tools field. Errors come back in the OpenAI error shape.

Can I list the models without a key?

Yes: curl https://api.openuncensored.com/v1/models returns every model id, free.

Do I need an account?

No. Buy a prepaid key on the credits page and use it as the API key. The key is the account, and it cannot spend more than you loaded.

Updated November 18, 2026. Scores measured November 8, 2026. Markdown version: /openai-compatible-uncensored-api.md