Ternary Bonsai 2 9B

Neural Network

Ternary Bonsai 2 9B is a compact open PrismML model for dialogue, reasoning, and function calling.

Main

/

Models

/

Text AI Models

/

Ternary Bonsai 2 9B

Specifications

Input
text, images, documents
Output
text
Context size
262 144 tokens
Max answer length
65 536 tokens
Reasoning effort
low, medium, high
Capabilities
reasoning, function calling
0,1 $

Prompt cost

(per 1M tokens)

0,16 $

Answer cost

(per 1M tokens)

*Prices are shown for API usage via ECO providers.
bothub
BotHub: Try neural networks for freebot

Caps remaining: 0 CAPS
Code example and API for Ternary Bonsai 2 9BWe offer full access to the OpenAI API through our service. All our endpoints fully comply with OpenAI endpoints and can be used both with plugins and when developing your own software through the SDK.Create API key
import OpenAI from 'openai';
const openai = new OpenAI({
  apiKey: '<your bothub access token>',
  baseURL: 'https://openai.bothub.chat/v1'
});


// Sync - Text generation 

async function main() {
  const chatCompletion = await openai.chat.completions.create({
    messages: [{ role: 'user', content: 'Say this is a test' }],
    model: 'ternary-bonsai-2-9b',
  });
} 

// Async - Text generation 

async function main() {
  const stream = await openai.chat.completions.create({
    messages: [{ role: 'user', content: 'Say this is a test' }],
    model: 'ternary-bonsai-2-9b',
    stream: true
  });

  for await (const chunk of stream) {
    const part: string | null = chunk.choices[0].delta?.content ?? null;
  }
} 
main();
illustaration

How it works Ternary Bonsai 2 9B?

Ternary Bonsai 2 9B is a compact open-weights language model from PrismML, designed for conversational products, reasoning, and function calling. It is based on the Qwen3.5 architecture, with weights stored in ternary format and packed into two bits, significantly reducing memory requirements during inference and simplifying deployment where every gigabyte counts. In BotHub, the model is available alongside others: switching does not require rewriting integrations, payment is in rubles based on token usage, and access works without a VPN or foreign card. For teams building multi-model systems, a unified OpenAI-compatible API is useful, while companies benefit from contracts, invoices, and limits in the admin panel. The model is suitable for chat assistants and internal support bots, step-by-step reasoning and task analysis, agents with external tool calling, prototypes on low-end hardware, and comparing responses with larger models directly in the BotHub interface.

Frequently asked questions about Ternary Bonsai 2 9B

You can use generated results for commercial purposes. You own all rights to the content you create. The only restriction: make sure your prompt does not include copyrighted third-party material. You are responsible for respecting the rights to any input data.

The model works with text, documents, and images: it parses file content, answers questions about them, reasons before answering, and calls external tools. It is suitable for document analysis, data extraction, writing and parsing code, as well as tasks requiring structured JSON output according to a specified schema.

In a single response, the model generates up to 65,536 tokens, and the total context for the request and response reaches 262,144 tokens. In practice, this means voluminous documents and long answers, but the exact size in words or pages depends on the language and formatting, so it is better to rely on the token counter.

Yes, the model can reason before answering: reasoning support is noted in our data. It builds intermediate steps and only then formulates a conclusion, which helps with mathematics, multi-condition tasks, code analysis, and logic verification. For simple queries, this is not required, and the answer can be obtained faster.

Yes, both capabilities are noted in our data. The model can access external functions and return responses in a structured format, such as JSON according to a specified schema. This is convenient for integrations: it decides which tool to call, and the result is easy to parse programmatically without manual editing.

The model accepts images and documents as input—this is noted in our data, so you can upload an image or file and ask questions about its content. Audio and video input is not noted in our data, so you should not count on it; check with support if this is important for your task.

There is no separate quality assessment for the Russian language in our data, so it would be dishonest to claim anything specific. Test the model with your own examples: provide typical questions, document fragments, and expected response formulations. This way, you will see if the accuracy is sufficient for your task before integrating it into your product.

Support ServiceOpen from 10:00 to 18:00 MSK