EN ▾
Get API key

OpenAI Alternative Models: One Uncensored Text API

Drop-in uncensored OpenAI API alternative

If you are comparing openai alternative models, start with what your current model refuses or costs you. This site offers one open-weight text model that answers without content refusals for lawful adult use, behind the OpenAI request format.

Updated

Several server towers linking to one highlighted node, representing a single-model alternative to OpenAI

Key points

The short answer

Among openai alternative models, this site offers exactly one: an open-weight text model, tuned to answer without content refusals for lawful adult use, served through the same HTTP interface the OpenAI SDKs already speak. You change the base URL and the key, set the model to uncensored, and your existing chat code runs.

That is a narrow offer on purpose. You do not get a catalogue of models, image generation, embeddings or fine-tuning. You get a text endpoint with streaming, tool calling and JSON mode, a 100,000-token context window, and billing at $0.15 per 1M input tokens and $0.50 per 1M output tokens. If that matches the job, switching takes minutes. If it does not, the sections below say where else to look.

One thing to be clear about early: this is not the official API of OpenAI, Meta, Mistral, Qwen, DeepSeek or any other vendor. The model is open-weight and is run by an independent provider. There is no routing across vendors behind the scenes.

Four kinds of alternative

When people say they want to replace OpenAI, they usually mean one of four different things. Each has a different cost in time and control.

  • Another hosted vendor with its own models. You pick a new SDK or a compatible endpoint, learn that vendor's quirks and accept its content policy. Check their current pricing before you compare numbers.
  • A router. One key, many vendors' models, and the freedom to swap per request. You still inherit the policy of whichever model you route to. We describe the difference on the router comparison page.
  • Self-hosting an open-weight model. You control everything, and you also own the GPUs, the inference server, the updates and the 3 a.m. alerts.
  • One hosted open-weight model. This is what the site offers. You get less choice than a router and far less work than self-hosting.

Most teams that land here do so for one reason: their application keeps getting refusals on content that is legal for adults, and rewording prompts has stopped working. A model tuned not to refuse solves that directly, and the OpenAI-style interface means no application rewrite.

Four factors decide which kind fits. Time: a compatible endpoint is the fastest to adopt and self-hosting the slowest. Control: self-hosting wins, routers come second. Policy: with a hosted vendor or router, the model's own content rules apply, so read them before you build on it. Cost: compare per-token prices on the vendors' own pages, because they change, and include your own engineering time when you price self-hosting. A rough rule is that if you would spend more than a week of engineer time on GPUs and tuning, a hosted endpoint is cheaper even at a high token volume.

What the endpoint gives you

Everything below is what the service does today. Anything not listed is not promised.

ItemValue
Base URLhttps://api.openaiapialternative.com/v1
Model IDuncensored
EndpointsPOST /v1/chat/completions and GET /v1/models
AuthAuthorization: Bearer <key>
Context window100,000 tokens, prompt plus completion
StreamingServer-sent events with stream: true, usage at the end
Tool callingtools, tool_choice, streamed tool_calls, results as role: "tool"
JSON moderesponse_format: {"type":"json_object"}
Samplingtemperature, top_p, stop, seed and the penalties are passed through
Price$0.15 per 1M input tokens, $0.50 per 1M output tokens
Failed requestsFailed or refused requests are not charged

Limits for a full key are 8 concurrent requests and 300 requests per minute. A trial key gets 2 concurrent and 60 per minute. The maximum body is 8 MB. The full numbers and error codes are on the docs page.

When a single uncensored model is the right pick

A model that does not refuse is useful when refusals are the failure you are fighting. Typical cases:

  • Fiction, roleplay and companion apps for adults, where the story needs dark or mature themes handled as writing.
  • Writing tools for novelists and screenwriters who need a collaborator that does not stop mid-scene.
  • Trust and safety work that needs to read, label or rewrite lawful but unpleasant text, for example to build test sets for a content filter.
  • Internal tools where a refusal from a general assistant breaks a pipeline and someone has to rerun it by hand.

The hard limit is the same everywhere: no sexual content involving minors, ever. Beyond that, you are responsible for the content your own users produce and for following the law where you operate. A model without refusals shifts moderation to you. If your product has public users, add your own filters on input and output.

A model that does not refuse still needs direction. Put the rules of your product in the system prompt: the tone, the point of view, what the character may and may not do, and when to stay out of a scene. In practice, a good system prompt of 300 to 600 words does more for output quality than any sampling setting, and it costs very little. At 500 tokens it adds $0.000075 per request on the input side, which is under a cent per hundred requests.

When another option fits better

Be honest with yourself about the parts of your stack this service does not cover.

  • You need image, audio or video generation. This is a text API only.
  • You need embeddings for search or retrieval. There is no embeddings endpoint, so keep that part where it is.
  • You need to fine-tune a model on your own data. Not offered.
  • You want to choose among many models per request. A router is the better tool.
  • Your rules require the data to stay on your own hardware. Self-hosting is the only way to guarantee that.

Plenty of teams run a mixed setup: a general model for ordinary tasks and this endpoint for the requests that others decline. Because both speak the same protocol, a small routing function in your code is enough.

Switching in practice

With the official Python SDK, three values change: the base URL, the key and the model name. Nothing else in your calls needs to move.

from openai import OpenAI

client = OpenAI(
    base_url="https://api.openaiapialternative.com/v1",
    api_key="YOUR_KEY",
)

reply = client.chat.completions.create(
    model="uncensored",
    messages=[
        {"role": "system", "content": "You are a co-writer for adult fiction."},
        {"role": "user", "content": "Continue the scene in the lighthouse."},
    ],
    temperature=0.9,
)
print(reply.choices[0].message.content)

The same call from the shell, which is a handy way to check that your key works:

curl https://api.openaiapialternative.com/v1/chat/completions \
  -H "Authorization: Bearer YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"uncensored","messages":[{"role":"user","content":"Say hello in five words."}]}'

Check the response headers as well. X-Balance-USD shows your remaining credit, and X-Request-Id is what you quote if you contact support.

What a trial run costs

Tokens are billed separately for input and output. Take a typical chat turn with 2,000 input tokens (a system prompt plus some history) and 500 output tokens. The input costs 2,000 / 1,000,000 x $0.15 = $0.0003. The output costs 500 / 1,000,000 x $0.50 = $0.00025. One turn is $0.00055, so 1,000 turns are $0.55.

The $0.50 trial credit therefore covers roughly 900 turns of that size. It is claimed with one click on the API key page after sign-up, needs no card, lasts 7 days and is limited to one per person. Sign-up works with Google or with email and password.

When you move to paid use, top-ups are any whole amount from $20 to $500, paid by crypto (USDT on TRC20, USDC on Base). There is a bonus of 5% from $50, 10% from $100 and 15% from $200, so a $100 top-up gives $110 of credit. Credit never expires and there is no subscription. See the pricing page for the table.

For a feel of monthly spend, imagine a roleplay app with 500 conversations per day. Each one sends 3,000 input tokens and gets 600 output tokens back. Daily input is 1.5M tokens, which costs 1.5 x $0.15 = $0.225. Daily output is 0.3M tokens, which costs 0.3 x $0.50 = $0.15. That is $0.375 per day and $11.25 over 30 days. The numbers shift with prompt length, but they show why a small team can run such a product on a modest balance.

A fair way to test it

Do not judge a model on one clever prompt. A short, repeatable test tells you more.

  1. Pull 50 real prompts from your logs, including the ones that produced refusals or rewrites.
  2. Run them through your current model and through uncensored with the same system prompt and settings.
  3. Count refusals, then read ten answers from each side by hand for tone and accuracy.
  4. Switch on JSON mode or tool calling if you use them, and check that the output parses.
  5. Replay a burst of traffic to see how the limits behave. The trial key allows 2 concurrent requests, so use a full key for load tests.

If the results hold, get a key on the API key page and keep the old provider as a fallback until you trust the new one.

Questions and answers

Is this the official OpenAI API?

No. It is a separate service that speaks the same request format. The model is an open-weight model run by an independent provider, and it is not the product of OpenAI or any other big vendor.

Can I use the OpenAI SDK without changes?

Yes, apart from the base URL, the API key and the model name. Set the model to uncensored and the usual chat completions calls, streaming and tool calling work.

Does it support images or embeddings?

No. It is text only, with no image, audio or video generation, no embeddings and no fine-tuning.

Are my prompts used for training?

No. Prompts are not used for training.

Your key is one form away

Create an account, copy the key, change the base URL. That is the whole setup.

Get API keyRead the docs