Gonka API Pricing & Supported Models

Gonka Broker provides access to open-source large language models and a text embedding model, billed in USD per 1M tokens with no subscription: you pay only for what you use.

Available models

The current models and their live prices are returned by the public GET /v1/models endpoint, no login required. They’re also shown on the Prices tab when you sign in to your dashboard.

Pricing fluctuates based on GPU supply and demand. The prices shown in your dashboard reflect the current rates.

Tip

When you top up your account, the current prices are locked for that deposit. Even if prices change later, your existing funds retain their original rates. See Billing & Deposits for details.

Caution

The model list is continuously expanding. The examples below may not reflect the current state. For the up-to-date list, query the GET /v1/models endpoint or check the Prices tab in your dashboard.

Sample models and pricing

Below are example models and rates to give you a sense of what’s available. These are approximate and subject to change.

Model	Input (per 1M tokens)	Output (per 1M tokens)
MiniMaxAI/MiniMax-M2.7	~$0.30	~$0.30
deepseek-ai/DeepSeek-V4-Flash-0731	~$0.25	~$0.25
moonshotai/Kimi-K2.6	~$0.35	~$0.35
BAAI/bge-m3 (embeddings)	~$0.01	—

For the most current pricing, always check your dashboard.

Reasoning behavior

Chat models handle reasoning (thinking) differently; whenever a model thinks, the thinking is billed as output tokens. What you can control differs per model:

Model	Thinking control
MiniMaxAI/MiniMax-M2.7	Always on; not controllable
deepseek-ai/DeepSeek-V4-Flash-0731	Off by default; can be turned on per request; reasoning_effort "high" / "max" = deepest reasoning
moonshotai/Kimi-K2.6	On by default; can be turned off per request

See reasoning control for the request parameters (all three common client formats are accepted) and where the reasoning appears in responses.

Embeddings model

BAAI/bge-m3 is a multilingual text embedding model for semantic search, RAG, clustering, and recommendations. It produces 1024-dimensional vectors, covers 100+ languages, and accepts inputs up to 8,192 tokens.

It is served through the OpenAI-compatible POST /v1/embeddings endpoint and billed on input tokens only; embeddings have no output tokens. See OpenAI API Compatibility for request parameters and examples.

Model naming

Use the full model identifier when making API requests. For example:

{
  "model": "MiniMaxAI/MiniMax-M2.7"
}

{
  "model": "moonshotai/Kimi-K2.6"
}

The exact model names are returned by GET /v1/models and shown on the Prices tab in your dashboard.