# Alibaba releases Qwen3 235B A22B Instruct model on Fireworks

> The new instruction‑tuned LLM has 235 B parameters (22 B active) and supports up to 1 M tokens, priced at $0.22 per input million and $0.88 per output million.

Oossa · 2026-10-06 · https://oossa.com/en/alibaba-releases-qwen3-235b-a22b-instruct-model-on-fireworks

Alibaba’s Qwen team has made its Qwen3‑235B‑A22B‑Instruct‑2507 model available through Fireworks AI. It is a mixture‑of‑experts language model with 235 billion total parameters but only 22 billion active per token. The model can handle up to a million tokens using Dual Chunk Attention, though most users are advised to stay under 16 K tokens for speed and cost. It runs in FP8 quantized format and supports streaming and tool‑calling via Qwen‑Agent.

## The facts

- 235 billion total parameters, 22 billion active per token
- $0.22 per 1 M input tokens, $0.88 per 1 M output tokens

## Why it matters

Developers can now build richer, longer‑context applications like complex reasoning or code assistance at a known per‑token cost.

## Sources & references

1. [Qwen3 235B A22B Instruct 2507](https://fireworks.ai/models/fireworks/qwen3-235b-a22b-instruct-2507) – Fireworks AI

Last updated: 2026-10-06