Alibaba Qwen: Qwen3 235B A22B Thinking 2507

Qwen3 235B A22B Thinking 2507 is Alibaba Qwen's open-weight mixture-of-experts reasoning model, released 2025-07 as the thinking-focused variant of the Qwen3-235B family. It activates 22B of its 235B parameters per forward pass and targets step-by-step reasoning, mathematics, science, and long-form generation. The model supports tool calling and agentic workflows across multilingual tasks.

Modalities
inTextoutText
Price / 1M−25%
$0.27 / $2.70$0.45 / $3.50
Context
128K
Released
Jul 25, 2025
Knowledge cutoff
Jun 30, 2025

Channels

Every channel serves the same model, but they differ in price and stability. On each API key in the Platform, you pick a primary and a backup channel per vendor, and each billing line shows which channel served the request.

Provider
Discount
Input /M
Output /M
Cache read /M
Cache write 5m /M
Cache write 1h /M
Defaultstable
−25%
$0.27
$2.70
—
—
—

Capabilities

ReasoningTool use

Parameters

Request parameters this model accepts through the API.

frequency_penaltyinclude_reasoningmax_tokenspresence_penaltyreasoningstoptemperaturetool_choicetoolstop_ktop_p

Cost calculator

Pick a channel and a usage scenario, then drag the sliders to match your workload. Each scenario assumes a different prompt-cache hit rate.

Channel
Scenario
Unit $0.27 / $2.70 /1MDaily chat: 40% cache hit rate assumed · cache read $0.03 /1M
List price $195.65On MirAPI$146.19/ moYou save $49.46Create an API key

Benchmarks

bar = score (0–100) · ▏ median

Scores come from vendor reports and public leaderboards, not private evals.

Intelligence Index
General ability
12.7
#109
Coding Index
Code
22.1
#71
Agentic Index
Agentic
1.3
#54
GPQA Diamond
Science QA
77.5
#83
τ-bench (airline)
Agentic tool use
62.0
#90

Showcase

Every chat model gets the same three fixed prompts. Outputs are archived verbatim, so you can copy a prompt and reproduce the result.

Task prompt · identical for all models
Build a single-file HTML page using three.js from a CDN: a low-poly planet with a tilted ring of ~200 orbiting particles, a soft key light plus rim light, slow auto-rotation, and drag-to-orbit controls. No build tools — one file only.
Archived output
Archive not published yet

Every model will run this exact prompt, and its output will appear here verbatim with token counts and the cost per run.

Quick start

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.mirapi.ai/v1",   # changed
    api_key=os.environ["MIRAPI_API_KEY"],   # changed
)

resp = client.chat.completions.create(
    model="qwen3-235b-a22b-thinking-2507",
    messages=[{"role": "user", "content": "Hello"}],
)
Wire Qwen3 235B A22B Thinking 2507 into your product

1M free tokens to start, no card required. It’s OpenAI-compatible, so you only change the base URL.

Start free