POST /v1/completions

The legacy OpenAI Completions API. You send a raw prompt and the model continues it, with no chat template around it. Any OpenAI SDK or tool works when you point its base URL at https://api.abliteratedllm.dev/v1. For conversations, instructions, tools, images or reasoning output, use POST /v1/chat/completions instead.

Placeholders: $API_BASE is https://api.abliteratedllm.dev/v1, $API_ORIGIN is https://api.abliteratedllm.dev, and $YOUR_KEY is your API key.

Request

curl $API_BASE/completions \
  -H "Authorization: Bearer $YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "glm-5.3-flash-abliterated",
    "prompt": "The lighthouse keeper opened the logbook and wrote:",
    "max_tokens": 256,
    "temperature": 0.8
  }'

Abridged response:

{
  "id": "…",
  "object": "text_completion",
  "created": 1793700000,
  "model": "glm-5.3-flash-abliterated",
  "choices": [
    { "index": 0, "text": " Day 41. The fog came in before the gulls did…", "finish_reason": "stop" }
  ],
  "usage": {
    "prompt_tokens": 12,
    "completion_tokens": 187,
    "total_tokens": 199,
    "prompt_tokens_details": { "cached_tokens": 0 }
  }
}

Not on Official DeepSeek API models

DeepSeek's API has no completions endpoint, so a request for a model labelled Official DeepSeek API (variant "official_api" in GET /v1/models) fails with 400 invalid_request before anything is charged. Every other model in GET /v1/models works here.

Parameters

Fields not listed here are passed to the model unchanged. Sampling parameters are range-checked first: a value outside its range fails with 400 invalid_request before anything is charged.

Streaming

With "stream": true the response is a stream of data: lines. Headers are sent as soon as the model accepts the request. Until the first token arrives, the stream carries an SSE comment line, : ping, every 15 seconds, so proxies don't close an idle connection. SSE clients and the OpenAI SDKs ignore comment lines.

: ping

data: {"id":"…","object":"text_completion","choices":[{"index":0,"text":" Day","logprobs":null,"finish_reason":null}]}

data: {"id":"…","object":"text_completion","choices":[{"index":0,"text":"","logprobs":null,"finish_reason":"stop"}]}

data: {"id":"…","object":"text_completion","choices":[],"usage":{"prompt_tokens":12,"completion_tokens":187,"total_tokens":199,"prompt_tokens_details":{"cached_tokens":0}}}

data: [DONE]

Each chunk carries the next piece of text for one choice, finish_reason arrives on a chunk of its own, and usage always arrives in a final chunk whose choices array is empty. Check that choices isn't empty before reading choices[0]:

from openai import OpenAI

client = OpenAI(base_url="$API_BASE", api_key="$YOUR_KEY")

stream = client.completions.create(
    model="glm-5.3-flash-abliterated",
    prompt="The lighthouse keeper opened the logbook and wrote:",
    max_tokens=256,
    stream=True,
)
for chunk in stream:
    if chunk.choices:
        print(chunk.choices[0].text, end="")
    if chunk.usage:
        print("\n", chunk.usage)

With an array prompt, each chunk belongs to the choice in its choices[].index, and chunks for different prompts can carry different id values.

Logprobs

An integer logprobs returns the legacy format on each choice: parallel arrays with one entry per output token.

{
  "tokens": [" Day", " 41"],
  "token_logprobs": [-0.21, -1.73],
  "top_logprobs": [{ " Day": -0.21, " The": -2.05 }, { " 41": -1.73, " 12": -1.94 }],
  "text_offset": [51, 55]
}

While streaming, each chunk carries the entries for its own tokens. The non-streaming response joins them per choice.

Context length

The prompt plus the output budget must fit in the model's context window, context_length in GET /v1/models. With the default "context_length_exceeded_behavior": "truncate", the output budget is lowered to the room the prompt leaves; with "error", such a request fails with 400 invalid_request. A prompt that is longer than the context window on its own always fails, and is never cut.

Prompt caching

Caching is automatic, as on chat completions: when a prompt starts with the same tokens as a recent one, those tokens are read from cache and billed at the cached-input price. usage.prompt_tokens_details.cached_tokens reports how many.

Billing

prompt_tokens includes cached tokens, which are billed at the cached-input price:

cost = (prompt_tokens - cached_tokens) × input price
     + cached_tokens × cached-input price
     + completion_tokens × output price

Prices are per 1M tokens and each term is rounded up to the next $0.000001. With "service_tier": "priority", the model's priority prices replace its standard ones. Current prices are on GET /v1/models, which also explains holds, aborted streams and failed requests.

Errors

Errors use the OpenAI shape, {"error": {"type", "code", "message"}}. The codes are on the Errors page. Every response carries x-request-id.