Qwen3 235B A22B Thinking 2507

Provider: Alibaba (Qwen) CN

Input price
$0.45
/ 1M tokens
Output price
$3.50
/ 1M tokens
Cache price
$0.1
/ 1M tokens (read)
Context
128K tokens
Max output 16K tokens

Overview

Alibaba's Qwen3 235B A22B Thinking 2507 is a reasoning-focused MoE model activating 22B of 235B parameters per pass for complex logic. It supports function-calling across a 128k context with no free tier.

Strengths

  • MoE architecture activating 22B parameters
  • Native function-calling capability integrated
  • Cached input priced at $0.10/1M
  • Knowledge cutoff date of June 2025

Best use cases

  • Complex multi-step algorithmic reasoning
  • Advanced mathematical problem solving
  • Structured data extraction pipelines

Pricing position

Input costs $0.45/1M and output costs $3.50/1M tokens, making output significantly pricier than peer models like GPT-4.1 Mini at $1.60/1M.

Caveats

Output pricing is high relative to siblings like Qwen3 Next 80B at $1.10/1M output.

Notes drafted by an LLM from the figures on this page (2026-11-10) and rewritten when the price changes.

Key facts

Price basis
Median third-party host price (no first-party API)
Max output
16K tokens
Knowledge cutoff
2025-06-30
Listed on OpenRouter
2025-07-25
Tracked since
2026-05-08
Last updated
2026-11-10

Features

Function callingReasoning

Monthly cost estimate

Scale Input (M) Output (M) Monthly
0.1M (hobby) 0.1 0.05 $0.22
1M (small app) 1 0.5 $2.20
10M (mid scale) 10 5 $22.00
100M (large) 100 50 $220.00

※ Cache and batch discounts not factored in. Detailed calculator

Official resources

Other models from Alibaba (Qwen)

Similar-priced models

Source: OpenRouter API (qwen/qwen3-235b-a22b-thinking-2507)