Qwen 3.8 Max 0902
Neural Network
Qwen 3.8 Max 0902 is an updated snapshot of Alibaba's Qwen 3.8 Max: a 2.4T parameter MoE model accepting text, images, and video.
Specifications
- Input
- text, images, documents
- Output
- text
- Context size
- 262 144 tokens
- Max answer length
- 131 072 tokens
- Reasoning effort
- minimal, low, medium, high, extra high
- Capabilities
- reasoning, function calling
Prompt cost
(per 1M tokens)
Answer cost
(per 1M tokens)
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: 'qwen3.8-max-0902:batch',
});
}
// Async - Text generation
async function main() {
const stream = await openai.chat.completions.create({
messages: [{ role: 'user', content: 'Say this is a test' }],
model: 'qwen3.8-max-0902:batch',
stream: true
});
for await (const chunk of stream) {
const part: string | null = chunk.choices[0].delta?.content ?? null;
}
}
main();
How it works Qwen 3.8 Max 0902?
Qwen 3.8 Max 0902 is an updated snapshot of the Qwen 3.8 Max model from the Alibaba Qwen team. It is a 2.4 trillion parameter Mixture-of-Experts model: it accepts text, images, and video as input and outputs text, allowing a single request to reference documents, images, and video sequences simultaneously. The 0902 snapshot is Qwen 3.8 Max fixed in a specific build, which is convenient when you need stable production responses and repeatable results. In BotHub, the model is available from Russia without a VPN or foreign card; payment is in rubles based on actual usage, tokens do not expire, and over 250 models are available in the same window. A unified OpenAI-compatible API allows switching Qwen to another model without rewriting code, while AES-GCM encryption and a no-data-retention policy address privacy concerns. The model is useful for parsing documents with diagrams and scans, describing video sequences in media and monitoring, acting as a multimodal support assistant, checking code via screenshots and logs, and analyzing tables and charts in research. For companies, we offer contracts, invoicing, EDI, an admin panel, and usage limits.Frequently asked questions about Qwen 3.8 Max 0902
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 accepts text, documents, and images, and returns text. It is suitable for working with code, parsing documentation, analyzing screenshots and tables, multi-step reasoning, and tasks involving external tools. A 262,144-token context allows loading large materials in their entirety without manual splitting.
In a single response, the model outputs up to 131,072 tokens—an impressive volume sufficient for long reports, detailed documentation, and large code files. This eliminates the need to stitch results together from fragments. The input context reaches 262,144 tokens, so source materials can be provided in full.
Yes, a reasoning mode is noted in the model's capabilities: it goes through intermediate steps before the final answer. This helps where logic is important—in mathematics, code analysis, planning, and resolving contradictory conditions. For simple queries, you can avoid increasing reasoning depth to get an answer faster.
Yes, both capabilities are confirmed. The model supports function calling, so it can be connected to external services and databases, and it also returns structured JSON output—convenient for parsing responses and automated pipelines. This makes it suitable for agentic scenarios and API integrations.
The model accepts images and documents: you can upload a scan, PDF, or screenshot and ask questions about the content. There is no mention of audio or video in our data, so we will not promise their processing—this point should be verified in the documentation. Generation of anything other than text is also not noted.
There are no separate measurements for the Russian language in our data, so we do not provide specific ratings. The easiest way to check quality is with your own tasks: take a few typical queries, compare the answers with a benchmark, and see if the precision of the wording is sufficient for your scenarios.