Browse all Qwen text models
Provider logo

Qwen 3 235b A22B 2507 Thinking

qwen/qwen3-235b-a22b-thinking-2507
Back
Provider logo

Qwen 3 235b A22B 2507 Thinking

qwen/qwen3-235b-a22b-thinking-2507
Back

The thinking version of Qwen 3 235b A22B 2507, with enhanced reasoning capabilities and step-by-step problem solving.

Added Sep 11, 2025

Model weights

Context Window

256.0K

Max Output

262.1K

Avg output tokens (7d)

1.6K tokens

86%

Input Price (Auto)

$0.30/1M

Output Price (Auto)

$0.50/1M

Benchmarks

Performance metrics and benchmarks

Sourced from Artificial Analysis.

Intelligence Index

12.7

Better than 49% of models compared

Coding Index

22.1

Better than 26% of models compared

Agentic Index

1.3

Better than 25% of models compared

Agentic work

AutomationBench-AA

Workflow automation with guardrail penalties

2.2%

Better than 21% of models compared

AA-Briefcase

Agentic knowledge work (Elo)

172 Elo

Better than 12% of models compared

GDPval-AA v2

Economically valuable tasks (Elo)

302 Elo

Better than 18% of models compared

Document reasoning

GDP.pdf

Professional PDF reasoning: all-pass rate

4.8%

Better than 23% of models compared

AA-LCR v1.1

Long context reasoning with updated grading

72.0%

Better than 67% of models compared

Reasoning

HLE

Humanity's Last Exam

15.9%

Better than 63% of models compared

IFBench

Instruction-following benchmark

51.2%

Better than 61% of models compared

CritPt

Research-level physics reasoning

0.0%

Coding

Terminal-Bench v4.0

Practical coding and terminal tasks

0.0%

Better than 16% of models compared

SciCode

Python programming for scientific computing

41.4%

Better than 30% of models compared

LiveCodeBench

Contamination-free coding benchmark

78.8%

Better than 90% of models compared

Math

AIME 2025

American Invitational Mathematics Examination 2025

91.0%

Better than 92% of models compared

AIME

American Invitational Mathematics Examination

94.0%

Better than 98% of models compared

Math-500

Diverse mathematical problem solving benchmark

98.4%

Better than 95% of models compared

Knowledge

MMLU-Pro

Professional and academic subject knowledge

84.3%

Better than 89% of models compared

AA-Omniscience Accuracy

Proportion of correctly answered questions

22.8%

AA-Omniscience Hallucination Rate

Rate of incorrect answers among non-correct responses

89.8%

Legacy benchmarks

GPQA Diamond (legacy)

Graduate-level scientific reasoning

79.0%

Better than 65% of models compared

Terminal-Bench Hard (legacy)

Agentic coding and terminal use

13.6%

Better than 50% of models compared

T²-Bench Telecom (legacy)

Conversational AI agents in dual-control scenarios

53.2%

Better than 54% of models compared

AA-LCR (unversioned / legacy)

Long context reasoning evaluation

72.0%

Better than 67% of models compared

GDPval-AA (unversioned / legacy)

Economically valuable tasks

0.0%

Last updated Oct 2, 2026

Artificial Analysis

Providers

Choose explicit providers for this model. Auto routing remains available as the default option.

Loading provider options…

Compare Qwen 3 235b A22B 2507 Thinking with similar models from the same provider or model family.

Qwen 3 235b A22B 2507

qwen/qwen3-235b-a22b-instruct-2507

Qwen 3 235b A22B Instruct 2507 the updated version of Qwen3 235B A22B, with significant improvements in performance. This model is non-thinking.

Qwen 3 235b A22B

qwen/qwen3-235b-a22b

Qwen 3 235b is a 235b model with 22B active parameters. Supports switching between thinking and non thinking: trigger thinking with /think and /no_think anywhere in a prompt or system message to toggle chain-of-thought reasoning.

Qwen3 VL 235B A22B Instruct

qwen/qwen3-vl-235b-a22b-instruct

Routed across multiple providers, which may include Alibaba, a Chinese entity; provider-specific privacy and logging guarantees may vary. Qwen3 Vision‑Language model (235B MoE, ≈22B active) tuned for instruction following and grounded visual QA. Excels at image understanding, dense OCR, charts/diagrams, and multi‑image context.

Qwen 3.8 27B Hemingway

qwen/qwen3.8-27b-hemmingway

Qwen 3.8 27B Hemingway is an open-weight NVFP4 multimodal creative finetune for long-form prose, character dialogue, storytelling, and roleplay.

Qwen 3.8 27B Cybersecurity

qwen/qwen3.8-27b-cybersecurity

Qwen 3.8 27B Cybersecurity is a cybersecurity-focused variant based on the uncensored model, with provider moderation for illegal activities. It supports optional reasoning, image understanding, tool calling, and a 262,144-token context window.

Qwen 3.8 27B Queen

qwen/qwen3.8-27b-queen

Qwen 3.8 27B Queen is an FP8 open-weight roleplay finetune with image understanding, tool calling, optional reasoning, and a 524,288-token context window.