api.cleus.ai · OpenAI-compatible

The uncensored AI API

One key for the Cleus chat models, abliterated DeepSeek, GLM, Qwen, Kimi and Hy4 models, 8 image and 8 video models, voices, music, characters and stories. Your system prompt goes through untouched, and you pay per use from a prepaid balance.

Needs Cleus Premium and a USD balance. Already have an account? Sign in.

6
abliterated open models
16
image and video models
58
REST endpoints
1M
token context on most chat models
Python · openai
from openai import OpenAI

client = OpenAI(
    base_url="https://api.cleus.ai/v1",
    api_key="clsk_live_...",
)

resp = client.chat.completions.create(
    model="cleus-smartest-3",
    messages=[
        {"role": "user", "content": "The reactor is leaking again."},
    ],
)
print(resp.choices[0].message.content)
Models
# the same key also takes the open models
model="deepseek-v4.1-flash-abliterated"
model="glm-5.3-abliterated"
model="glm-5.3-flash-abliterated"
model="qwen-3.8-27b-abliterated"
model="kimi-k3.1-pro-abliterated"
model="hy4-abliterated"
What is in it

One API, every kind of output

Chat is OpenAI-compatible. Everything else is plain REST with JSON in and URLs out, described in an OpenAPI 3.1 spec.

Uncensored chat and reasoning

The Cleus Fast and Smartest models, and abliterated open models under their own names. Streaming, OpenAI-compatible, 1M-token context on most.

Images: 8 models and every edit

Text to image, multi-image edits, 4× upscale, background removal, inpainting, outpainting, try-on, 3D models and SVG.

Video: 8 models, edits and swaps

Text and image to video up to 30 s at 1080p with sound, video edits, swaps, extends, talking videos, lip sync, dubbing and 4k upscale.

Voices and music

Text to speech, voice cloning from a short sample, voice changing, sound effects, and full songs with vocals and cover art.

Characters and stories

Search, create and chat with roleplay characters that remember, and write long stories chapter by chapter with story memory.

Agents, files and jobs

The Cleus Agent researches the web, runs code and makes files. The Imagine Agent directs whole productions. One endpoint polls every slow job.

Cleus models

Cleus chat models

Cleus’s own uncensored models, the same ones as chat on cleus.ai. They answer as Cleus, and the agent can search the web, run code and make images and video inside a reply.

Model idWhat it isPrice / 1M tokens
cleus-agentAgentic model with tool use: web search, image generation, code execution.$0.02 a reply + tools
cleus-fastFast uncensored general-purpose chat model.$0.99 in · $3.96 out
cleus-fast-v2Fast uncensored chat model with the fewest refusals.$0.60 in · $2.70 out
cleus-fast-v3Fast uncensored chat model that answers without a thinking pause.$0.99 in · $3.96 out
cleus-smartest-1Deep reasoning model for complex problems.$1.50 in · $6.45 out
cleus-smartest-2Latest flagship reasoning model.$1.95 in · $10.23 out
cleus-smartest-3Deep uncensored reasoning model with the fewest refusals.$0.99 in · $3.96 out
cleus-coding-3Uncensored coding and reasoning model for code, agents and security work. Text only.$3.00 in · $6.00 out
Open models

Abliterated open models, under their own names

Open-weight models with the refusals removed, served as a direct line: your system prompt, tool calls and sampling settings reach the model as you sent them. Each has its own page with specs, code and pricing.

DeepSeek · abliterated

DeepSeek V4.1 Flash Abliterated

deepseek-v4.1-flash-abliterated

DeepSeek V4.1 Flash with refusals removed. Thinking on or off, tool calls, images, 1M context.

1M contextText + imagesTool callsThinking on/off
$1.09 / $4.36per 1M tokens, in / out
DeepSeek V4.1 Flash Abliterated API →
Z.ai · abliterated

GLM 5.3 Abliterated

glm-5.3-abliterated

GLM 5.3 with refusals removed. A large coding and reasoning model: always thinks, tool calls, text only, 1M context.

1M contextText onlyTool callsAlways thinks
$3.30 / $6.60per 1M tokens, in / out
GLM 5.3 Abliterated API →
Z.ai · abliterated

GLM 5.3 Flash Abliterated

glm-5.3-flash-abliterated

GLM 5.3 Flash with refusals removed. Cheap and quick for everyday, high-volume work: tool calls, images, 1M context.

1M contextText + imagesTool callsAlways thinks
$0.40 / $1.65per 1M tokens, in / out
GLM 5.3 Flash Abliterated API →
Alibaba · abliterated

Qwen 3.8 27B Abliterated

qwen-3.8-27b-abliterated

Qwen 3.8 27B with refusals removed. The cheapest input here: thinking on or off, tool calls, images, 262K context.

256K contextText + imagesTool callsThinking on/off
$0.13 / $4.46per 1M tokens, in / out
Qwen 3.8 27B Abliterated API →
Moonshot AI · abliterated

Kimi K3.1 Pro Abliterated

kimi-k3.1-pro-abliterated

Kimi K3.1 Pro with refusals removed. The strongest model here, for the hardest coding, agent and security work: thinking on or off, tool calls, images, 1M context.

1M contextText + imagesTool callsThinking on/off
$6.50 / $19.50per 1M tokens, in / out
Kimi K3.1 Pro Abliterated API →
Tencent · abliterated

Hy4 Abliterated

hy4-abliterated

Tencent Hy4 with refusals removed. A mid-priced all-rounder that reads images: thinking on or off, tool calls, 1M context.

1M contextText + imagesTool callsThinking on/off
$2.08 / $4.16per 1M tokens, in / out
Hy4 Abliterated API →
Media models

8 image and 8 video models

Used by /images/generations, /images/edits and /videos/generations. The older ids cleus-image-quality, cleus-image-creative and cleus-video-quality keep working. GET /v1/models returns every model with its limits and exact prices.

Model idKindWhat it isFrom
uncensoredImageUncensored+: No filter, allows explicit (adults only), realistic photos and readable text$0.09 / image
qualityImageQuality+: Fast, cheap and sharp; good default; allows explicit$0.09 / image
proImagePro+: Most detailed and photoreal, up to 2K$0.36 / image
nano-banana-2.1ImageNano Banana 2.1: Precise edits, readable text, character consistency$0.12 / image
gpt-image-2.5ImageGPT Image 2.5: Posters, logos, UI mockups, typography$0.27 / image
grok-imagineImageGrok Imagine 2.0: Real public figures and celebrities$0.24 / image
nano-banana-proImageNanoBanana Pro: Careful, safe-for-work edits$0.09 / image
flux-2ImageFlux 2: Cheap square images and quick edits$0.06 / image
qualityVideo · 2–30 sQuality+: Best all-rounder: 2–30 s, up to 1080p, native sound and dialogue, start + end frames$0.69 / clip
uncensoredVideo · 5 or 10 sUncensored+: No filter, allows explicit (adults only), 5 or 10 s, optional background music$0.51 / clip
klingVideo · 3–30 sKling 4.0: Cinematic motion with sound, 3–30 s, 720p or 1080p, start + end frames$1.58 / clip
seedanceVideo · 4–15 sSeedance 2.5: Top detail, multi-shot scenes and dialogue, 4–15 s$2.22 / clip
seedance-2Video · 3–12 sSeedance 2.0: Reliable motion with native audio, 3–12 s, 720p, start + end frames$1.80 / clip
veoVideo · 4 or 8 sVeo 3: Realistic footage with sound, 4 or 8 s; can be extended with cleus_extend_video$1.44 / clip
soraVideo · 4 or 8 sSora 2: Storytelling shots with sound, 4 or 8 s$1.20 / clip
grok-imagineVideo · 1–15 sGrok Imagine Video: Real public figures, 1–15 s, up to 1080p, with sound$0.84 / clip
Quickstart

First request in three steps

Chat follows the OpenAI format. If you already have OpenAI code, change two lines: the base URL and the key.

1. Create a key

Sign in, open this page and create an API key. It starts with clsk_live_ and is shown once, so copy it then. You can revoke a key at any time.

2. Add balance

The API needs Cleus Premium and a USD balance. Every request is paid from the balance as it runs. Nothing is billed afterwards.

3. Send a request

Use the base URL https://api.cleus.ai/v1 and send the key as a Bearer token. Pick any model id from the tables above.

cURL
curl https://api.cleus.ai/v1/chat/completions \
  -H "Authorization: Bearer $CLEUS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4.1-flash-abliterated",
    "messages": [
      {"role": "system", "content": "You are a blunt code reviewer."},
      {"role": "user", "content": "Review this function: ..."}
    ]
  }'
Python · openai
# pip install openai
from openai import OpenAI

client = OpenAI(
    base_url="https://api.cleus.ai/v1",
    api_key="clsk_live_...",
)

resp = client.chat.completions.create(
    model="deepseek-v4.1-flash-abliterated",
    messages=[{"role": "user", "content": "Hello!"}],
)
print(resp.choices[0].message.content)
Node.js · openai (streaming)
// npm install openai
import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.cleus.ai/v1",
  apiKey: process.env.CLEUS_API_KEY,
});

const stream = await client.chat.completions.create({
  model: "cleus-smartest-3",
  messages: [{ role: "user", content: "Hello!" }],
  stream: true,
});
for await (const chunk of stream) {
  process.stdout.write(chunk.choices[0]?.delta?.content ?? "");
}
Guide

Authentication and limits

Every request carries your key in the Authorization header.

Header. Authorization: Bearer clsk_live_… on every request. Keys belong to your account: anyone with the key can spend your balance, so keep it on a server and out of client-side code.

Access. Calls that cost money need Premium and a balance above zero. Read-only calls such as GET /v1/models, GET /v1/balance and GET /v1/jobs/{id} only need a valid key.

Rate limit. 90 requests a minute per account, across all keys. Over the limit you get 429 with a Retry-After header in seconds.

Keys. Only a hash of each key is stored, which is why a key cannot be shown again. Revoking a key stops it at once.

Check your key and balance
curl https://api.cleus.ai/v1/balance \
  -H "Authorization: Bearer $CLEUS_API_KEY"

# → {"object": "balance", "balance_cents": 1234, "balance_usd": 12.34}
Guide

Chat completions

POST /v1/chat/completions serves two families of models. Both return the OpenAI response shape.

Cleus models: the Cleus assistant

cleus-smartest-3, cleus-fast-v3, cleus-agent and the rest. They answer as Cleus through the same pipeline as chat on cleus.ai: system messages are read as instructions inside the prompt, and tool calls and sampling settings are not used. Best when you want the Cleus assistant or the agent’s built-in tools.

Open models: a direct line

deepseek-v4.1-flash-abliterated, glm-5.3-abliterated, glm-5.3-flash-abliterated, qwen-3.8-27b-abliterated, kimi-k3.1-pro-abliterated, hy4-abliterated. Messages keep their roles, your system prompt is the only system prompt, and tools, sampling settings and thinking effort are passed to the model. Replies stream straight from the model. Best for your own apps, characters and agents.

FieldModelsWhat it does
modelallA chat model id from the tables above.
messagesallOpenAI-style messages. Open models get them as sent. Cleus models read system messages as instructions inside the prompt.
streamalltrue for server-sent events.
max_tokensopen modelsUpper limit for the reply, thinking included. max_completion_tokens is accepted too.
temperature, top_p, stopopen modelsSampling settings, passed to the model. Some models also take top_k, seed and penalties: see each model page.
tools, tool_choiceopen modelsOpenAI-style function tools. The reply carries tool_calls.
response_formatopen modelsPassed to the model for JSON output.
reasoning_effortopen modelsHow long the model thinks: none, low, medium, high or max, depending on the model. {"reasoning": {"effort": "…"}} works too.

Streaming. With "stream": true you get OpenAI-style chunks. While a model thinks, its thinking arrives as delta.reasoning_content, and quiet stretches carry : keep-alive comments so proxies keep the connection open. Open models send token usage in the last chunk.

Thinking. Reasoning models think before they answer, and thinking is billed as output. On open models set reasoning_effort to trade speed for depth; "none" switches thinking off where the model allows it.

Long replies. A Cleus model that is still writing after about five minutes hands the reply back as a job: a non-streamed request returns 202 with an id, a stream ends with finish_reason: "length" and cleus_job.id. Fetch the finished reply from GET /v1/jobs/{id}.

Output. Text comes back as Markdown in choices[0].message.content, with code in fenced blocks. The agent adds media links in the text and in message.cleus_media.

A stream
data: {"choices":[{"delta":{"role":"assistant"}}]}
data: {"choices":[{"delta":{"reasoning_content":"The user wants…"}}]}
data: {"choices":[{"delta":{"content":"Here is"}}]}
data: {"choices":[{"delta":{"content":" the answer."}}]}
: keep-alive
data: {"choices":[{"delta":{},"finish_reason":"stop"}],"usage":{…}}
data: [DONE]
The agent: research, then a poster
curl https://api.cleus.ai/v1/chat/completions \
  -H "Authorization: Bearer $CLEUS_API_KEY" -H "Content-Type: application/json" \
  -d '{
    "model": "cleus-agent",
    "messages": [{"role": "user", "content": "Search the latest SpaceX news and make me a poster of it."}]
  }'

# → message.content holds Markdown with the image link,
#   message.cleus_media = {"images": ["https://…"], "video": null, "audio": null}
Guide

Images, video and audio

Results come back as hosted URLs. Download what you want to keep: links are not permanent.

Generate an image
curl https://api.cleus.ai/v1/images/generations \
  -H "Authorization: Bearer $CLEUS_API_KEY" -H "Content-Type: application/json" \
  -d '{"model": "cleus-image-quality", "prompt": "a neon city at dusk", "n": 1}'

# → {"id": "-Nx…", "model": "cleus-image-quality", "data": [{"url": "https://…/image.webp"}]}
Edit an image
curl https://api.cleus.ai/v1/images/edits \
  -H "Authorization: Bearer $CLEUS_API_KEY" -H "Content-Type: application/json" \
  -d '{"model": "cleus-image-creative", "prompt": "make it snow", "image": "https://…/photo.png"}'

# image: a URL, base64 data, or a multipart upload
Avatars: the same face in any image
# 1) once: make an Avatar from 3–10 photos of one person (free)
curl https://api.cleus.ai/v1/avatars \
  -H "Authorization: Bearer $CLEUS_API_KEY" -H "Content-Type: application/json" \
  -d '{"name": "Maya", "adults_confirmed": true, "wait": 90,
       "photo_urls": ["https://…/1.jpg", "https://…/2.jpg", "https://…/3.jpg"]}'
# → {"object": "avatar", "status": "completed", "avatar_id": "api3f9c…", "name": "Maya"}
#   GET /v1/avatars lists the ones you already have, with the same ids

# 2) any time, anywhere: new photos of them
curl https://api.cleus.ai/v1/images/generations \
  -H "Authorization: Bearer $CLEUS_API_KEY" -H "Content-Type: application/json" \
  -d '{"avatar_id": "api3f9c…", "prompt": "hiking in Patagonia at golden hour", "n": 2}'
# → {"id": "-Nx…", "model": "cleus-avatar", "data": [{"url": "https://…"}, {"url": "https://…"}]}

An Avatar is a saved face with an id. Make one from 3–10 photos of one person, through the API or on the Avatars page, then send its avatar_id to /images/generations from any app, script or agent. The face stays the same in every scene.

Where the id is. GET /v1/avatars returns every Avatar on your account with its id, name, status and a preview. Each Avatar's page in the app shows the id with a copy button too.

Rules. avatar_id replaces model, takes n up to 4 and an aspect_ratio, and only works with the key of the account that owns the Avatar. Everyone in the photos must be 18 or older and have agreed to it, and Avatar images are never explicit.

Generate a video, then poll the job
# 1) start the job
curl https://api.cleus.ai/v1/videos/generations \
  -H "Authorization: Bearer $CLEUS_API_KEY" -H "Content-Type: application/json" \
  -d '{"model": "kling", "prompt": "slow dolly-in on a neon samurai, rain", "duration": 5, "resolution": "720p"}'
# → 202 {"id": "-Oq7x…", "status": "processing"}

# 2) poll until it is done (or add ?wait=60 to wait up to 60 s)
curl https://api.cleus.ai/v1/jobs/-Oq7x… -H "Authorization: Bearer $CLEUS_API_KEY"
# → {"id": "-Oq7x…", "status": "completed", "data": [{"url": "https://…/video.mp4"}]}
Speech in a cloned voice
curl https://api.cleus.ai/v1/audio/voice-clone \
  -H "Authorization: Bearer $CLEUS_API_KEY" -H "Content-Type: application/json" \
  -d '{"text": "Welcome back, traveler.", "voice_sample_url": "https://…/me.wav"}'

Slow work is a job. Videos, songs and long chapters answer 202 with an id. Poll GET /v1/jobs/{id}, or pass "wait": 60 in the request to hold the connection for up to that many seconds.

Inputs. Wherever an endpoint takes an image, video or audio file you can send a URL, or upload first with POST /v1/files and use the link it returns.

Price first. Call a generation tool through POST /v1/tools/{name} with "get_cost": true to get its price without running anything.

Guide

Characters and stories

The roleplay characters and the story writer from the Cleus app, with their memory, over REST. What you create shows up in your Cleus account too.

Chat with a character
curl https://api.cleus.ai/v1/characters/l-lawliet/messages \
  -H "Authorization: Bearer $CLEUS_API_KEY" -H "Content-Type: application/json" \
  -d '{"message": "Who do you think Kira is?", "persona": {"name": "Light"}}'
Start a story
curl https://api.cleus.ai/v1/stories \
  -H "Authorization: Bearer $CLEUS_API_KEY" -H "Content-Type: application/json" \
  -d '{"title": "The Glass Orchard", "plot": "A thief steals memories…", "illustrate": true}'
Reference

Errors and status codes

Errors use the OpenAI shape, so SDK error handling works unchanged.

StatusTypeMeaning
200OKThe result, or a stream of chunks.
202Still runningA slow job. Poll GET /v1/jobs/{id} with the id in the body.
400invalid_request_errorBad JSON, a missing field, an unknown model, or a value the model does not accept.
401authentication_errorMissing, wrong or revoked API key.
402subscription_required / insufficient_balanceThe account has no Premium, or the balance is empty.
403permission_errorThe account that owns the key is disabled.
404not_foundNo such job, character, story or file.
429rate_limit_errorMore than 90 requests in a minute, or the model is busy. Wait for Retry-After.
502api_errorThe model failed to answer. Not billed. Retry.
An error
HTTP/1.1 402
{
  "error": {
    "message": "Insufficient balance. Add USD balance at https://cleus.ai/add-credits to use the API.",
    "type": "insufficient_balance",
    "code": null
  }
}
Reference

Every endpoint

All paths are under https://api.cleus.ai/v1. Each one mirrors a tool of the Cleus MCP server, so your code and your coding agent get the same features.

Text

  • POST/v1/chat/completionsOpenAI-compatible chat with the uncensored and reasoning models (streaming).
  • POST/v1/agentHand a task to the Cleus Agent: web research, files, code, ~1,450 specialist models.

Images

  • POST/v1/images/generationsGenerate images with any of the 8 image models, or of one of your Avatars with avatar_id (OpenAI-shaped response).
  • POST/v1/images/editsEdit or combine up to 4 images by instruction (OpenAI-shaped response).
  • POST/v1/images/upscaleUpscale an image 4× with face enhancement.
  • POST/v1/images/remove-backgroundCut the subject out onto a transparent PNG.
  • POST/v1/images/inpaintRepaint the white area of a mask.
  • POST/v1/images/expandOutpaint beyond an image edge.
  • POST/v1/images/try-onPut a garment from one photo on the person in another.
  • POST/v1/images/describeDetailed description / alt text for an image.
  • GET/v1/avatarsYour Avatars — use one as avatar_id in /images/generations.
  • POST/v1/avatarsMake an Avatar from 3–10 photos of one person (free, about a minute).
  • POST/v1/images/3dTurn an image of one object or character into a GLB 3D model.
  • POST/v1/images/vectorizeConvert an image to SVG, or generate SVG logos and icons.
  • GET/v1/templatesOne-tap photo and video effect templates.
  • POST/v1/templates/runRun an effect template on your photos.

Video

  • POST/v1/videos/generationsText- or image-to-video with any of the 7 video models (async job).
  • POST/v1/videos/editsRestyle/edit a video, or follow a reference clip (Kling 4.0).
  • POST/v1/videos/swapPut the person from a photo into a video’s exact motion.
  • POST/v1/videos/extendExtend a Veo video by ~7 seconds.
  • POST/v1/videos/talkingA portrait speaks a script (21 voices) or your audio — lip-synced, up to 60 s.
  • POST/v1/videos/lip-syncRe-sync a video’s lips to new audio (dubbing, new lines).
  • POST/v1/videos/enhanceUpscale to 4k, transparent background, or slow motion.
  • POST/v1/videos/analyzeShot-by-shot breakdown, cuts and a timed transcript of a video.
  • POST/v1/videos/facelessOne idea → narrated, captioned faceless short videos, 1–4 parts (async).
  • POST/v1/videos/remixClone a reference video’s format, swap someone in, or restyle it (async).
  • POST/v1/videos/dubDub a video into another language in the speakers’ own voices.
  • POST/v1/videos/reframeChange a video’s aspect ratio by generating the missing edges.
  • POST/v1/videos/soundSynced sound effects and ambience for a video.
  • POST/v1/imagine-agentImagine Agent: plans, generates, reviews and edits whole productions (async).

Audio

  • POST/v1/audio/speechText to speech with preset voices.
  • POST/v1/audio/voice-cloneSpeak text in a voice from a short sample.
  • POST/v1/audio/musicCompose a full song with vocals and cover art, or an instrumental.
  • POST/v1/audio/sound-effectsA sound effect or ambience from a description, up to 22 s.
  • POST/v1/audio/voice-changeRe-voice speech in another voice, same words and delivery (audio or video).
  • GET/v1/audio/voicesPreset voices and your saved voice clones.

Characters

  • GET/v1/charactersSearch public characters (or ?mine=true for yours).
  • POST/v1/charactersCreate a character.
  • GET/v1/characters/{id}Character card, optionally with ?history=N recent messages.
  • POST/v1/characters/{id}/messagesSend a message and get the in-character reply.

Stories

  • GET/v1/storiesList your stories.
  • POST/v1/storiesStart a story and write chapter 1.
  • GET/v1/stories/{id}Premise, cast, chapter list and chapter text (?chapter=N, 0 = all).
  • PATCH/v1/stories/{id}Change settings for the chapters ahead.
  • POST/v1/stories/{id}/chaptersWrite the next chapter, optionally steered.
  • PATCH/v1/stories/{id}/chapters/{n}Rename, rewrite, delete or move a chapter (action: rename | edit | delete | move).

Files & jobs

  • POST/v1/filesUpload a file (multipart, base64 or URL) or get a signed upload URL.
  • GET/v1/generationsYour recent images, videos and audio.
  • GET/v1/jobs/{id}Status and result of any job (videos, songs, chapters, replies…).
  • GET/v1/toolsEvery tool with its JSON schema (same as the MCP server).
  • POST/v1/tools/{name}Call any tool by name with its JSON arguments.
  • GET/v1/assetsYour @tag asset library (POST with action "save" to add one).

Account

  • GET/v1/modelsAll chat, image, video and audio models.
  • GET/v1/models/searchSearch 1,400+ more models (?query=…), or ?model=<id> for its inputs and price.
  • POST/v1/models/runRun any of those models with its own inputs, priced up front.
  • GET/v1/balanceYour remaining USD balance.
  • GET/v1/transactionsBalance, recent top-ups and spend per day.

Security

  • POST/v1/security/analyzeOffline and passive security checks: JWT, hashes, SPF/DKIM/DMARC, CVEs, TLS, recon.

Full schemas: https://api.cleus.ai/v1/openapi.json (OpenAPI 3.1) · every tool by name: GET /v1/tools, POST /v1/tools/{name}

Pricing

Pay for what you use

Everything is paid from a prepaid USD balance at the moment the work is done. No invoices, no minimum spend on the API, and failed jobs are not charged.

Cleus Fast$0.99 in / $3.96 out / 1M tokens
Cleus Fast v2$0.60 in / $2.70 out / 1M tokens
Cleus Fast v3$0.99 in / $3.96 out / 1M tokens
Cleus Smartest 1$1.50 in / $6.45 out / 1M tokens
Cleus Smartest 2$1.95 in / $10.23 out / 1M tokens
Cleus Smartest 3$0.99 in / $3.96 out / 1M tokens
Cleus Coding 3$3.00 in / $6.00 out / 1M tokens
DeepSeek V4.1 Flash Abliterated$1.09 in / $4.36 out / 1M tokens
GLM 5.3 Abliterated$3.30 in / $6.60 out / 1M tokens
GLM 5.3 Flash Abliterated$0.40 in / $1.65 out / 1M tokens
Qwen 3.8 27B Abliterated$0.13 in / $4.46 out / 1M tokens
Kimi K3.1 Pro Abliterated$6.50 in / $19.50 out / 1M tokens
Hy4 Abliterated$2.08 in / $4.16 out / 1M tokens
Cleus Agent$0.02 / reply + each tool it uses
Web search+$0.01 / search
Images (by model)$0.06–$0.36 / image
Image edits (by model)$0.12–$0.36 / edit
Video (by model and resolution)from $0.51 / 5 s clip
Speech and voice cloning$0.08 / 1,000 characters
Music$0.12 / song
Character replies$0.02 / reply
Story chapters$6 / 1M tokens
Upscale, describe, uploads, readsFree

Chat is billed per token. You pay for the tokens the model reads and writes, at the price of the model you called. Thinking counts as output. Charges are exact to a fraction of a cent: the balance moves in whole cents and the remainder carries to your next request.

Open models count exactly what you sent. Cleus models also count Cleus’s built-in instructions as input.

The agent is billed per action. $0.02 for the reply, plus each tool it uses at that tool’s normal price: a web search, an image, a video. A step that fails is not charged.

Media has a fixed price per result, set by model, length and resolution. GET /v1/models lists the exact figures.

How a chat request is priced
cost_usd = input_tokens  / 1_000_000 * input_price
         + output_tokens / 1_000_000 * output_price

# deepseek-v4.1-flash-abliterated: 2,000 tokens in, 800 out
#   0.002 * $1.09 + 0.0008 * $4.36 = $0.0057
FAQ

Questions about the Cleus API

What is the Cleus API?

One API for uncensored AI: the Cleus chat and reasoning models, abliterated open models under their own names, image and video generation and editing, speech, voice cloning, music, roleplay characters and long-form stories. Chat follows the OpenAI format, so existing OpenAI code works after changing the base URL and key.

Is the API really uncensored?

The chat models are built on abliterated weights, which means the refusal behaviour is removed from the model itself. The open models go a step further: your request reaches the model exactly as sent, with no Cleus system prompt. The Cleus terms still apply to how you use the output.

What is the difference between Cleus models and open models?

Cleus models (cleus-smartest-3, cleus-fast-v3 and the rest) answer as Cleus and run through the same pipeline as chat on cleus.ai. Open models (DeepSeek V4.1 Flash Abliterated, GLM 5.3 Abliterated, GLM 5.3 Flash Abliterated, Qwen 3.8 27B Abliterated, Kimi K3.1 Pro Abliterated and Hy4 Abliterated) are a direct line to the model: your system prompt, tool calls and sampling settings pass through untouched.

What do I need to start?

A Cleus account with Premium and a USD balance. Create an API key on this page, then send requests with the key as a Bearer token. Usage is taken from the balance as you go.

How is usage billed?

From a prepaid USD balance, never after the fact. Chat is billed per token, at the price listed for each model. Images, videos, speech and music have a fixed price per result. Charges are exact to a fraction of a cent, and failed jobs are not charged.

Does it work with the OpenAI SDK, LangChain, SillyTavern or Open WebUI?

Yes. Anything that accepts an OpenAI-compatible base URL works: set the base URL to https://api.cleus.ai/v1, use your Cleus key, and pick a model id from the list.

Can I use Cleus inside Claude Code, Codex or Cursor?

Yes, through the Cleus MCP server at api.cleus.ai/mcp, which exposes the same features as tools. There is also Cleus Code, a desktop and terminal coding agent that runs on the same balance.

What are the rate limits?

90 requests a minute per account. Going over returns a 429 with a Retry-After header. Email founder@cleus.ai if you need more.

Start building with Cleus

Create a key and send your first request in a minute.