Large Language Models

Guides for using Bitdeer AI's text generation models — capabilities, parameters, and advanced features.

Model capabilities

Large Language Models (LLMs) on Bitdeer AI are accessed through the Chat Completions HTTP API (POST /v1/chat/completions). They support:

Text Generation — Generate coherent, contextual text for any task
Code Generation — Write, explain, and debug code in Python, JavaScript, and 30+ languages
Summarization & Translation — Condense documents or translate between languages
Multi-turn Conversation — Maintain context across multiple user/assistant exchanges
Tool Use — Call external functions and APIs via Function Calling

Available models

Model	Provider	Modality	Highlights
`deepseek-ai/DeepSeek-V4.1-Flash`	DeepSeek	Text	Multimodal mixture-of-experts
`deepseek-ai/DeepSeek-V4-Flash`	DeepSeek	Text	High-speed text generation
`zai-org/GLM-5.3`	ZAI	Text	Advanced reasoning and scaling
`zai-org/GLM-5.3-Flash`	ZAI	Text + Vision	Fast multimodal reasoning
`Qwen/Qwen3.8-27B`	Alibaba	Text	Compact dense vision-language

See the full list in the Model Catalog.

API integration

Use the API Reference for request and response fields. In short:

Base URL: https://api-inference.bitdeer.ai/v1
Send Authorization: Bearer <your Bitdeer key>

The snippet below uses the Python helper from SDKs; the same call is a JSON POST over HTTPS.

from openai import OpenAI

client = OpenAI(
    base_url="https://api-inference.bitdeer.ai/v1",
    api_key="YOUR_BITDEER_API_KEY",
)

response = client.chat.completions.create(
    model="deepseek-ai/DeepSeek-V4.1-Flash",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "What is serverless inference?"},
    ],
    stream=True,
    max_tokens=512,
)

for chunk in response:
    print(chunk.choices[0].delta.content or "", end="")

Feature guides

Streaming

Receive tokens in real time via Server-Sent Events.

Vision

Send images alongside text to vision-language models.

Audio

Send audio via input_audio for transcription and analysis.

Function Calling

Let models invoke external tools and APIs.

Structured Outputs

Force model responses to conform to a JSON Schema.

Key parameters

Parameter	Type	Description
`model`	string	Required. The model ID from the Model Catalog.
`messages`	array	Required. The conversation history — `system`, `user`, and `assistant` messages.
`max_tokens`	integer	Maximum tokens in the response.
`temperature`	float	Controls randomness. `0` = deterministic, `2` = highly creative. Default `1`.
`top_p`	float	Nucleus sampling. Controls cumulative probability. Default `1`.
`stream`	boolean	Whether to stream the response. Default `false`.
`stop`	string or array	Sequences that will terminate generation.
`frequency_penalty`	float	Penalizes repeated tokens. Range `[-2, 2]`.
`presence_penalty`	float	Encourages topic diversity. Range `[-2, 2]`.