Language modelActive

Qwen 3 235B A22B Thinking 2507

Alibaba

Released: July 25, 2025
Data date: September 8, 2026

Speed: 54.8 tokens/s
Output throughput
Price: $0.3 · $3
Input · Output / 1M tokens
Input: Text
Output: Text

Qwen 3 235B A22B Thinking 2507 is the open reasoning checkpoint in the large 235B class. It activates 22 billion of 235 billion parameters and produces additional internal reasoning in Thinking mode. Its context window is 262,144 tokens, with possible extension to 1,010,000.

The 2507 label identifies a separate reasoning checkpoint. Qwen 3 235B A22B Instruct 2507 instead gives direct answers without thinking mode. The original Qwen 3 235B A22B still allowed switching between both modes within the same checkpoint.

Specifications and access

Specification	Value and source
Weights	Open weightsSource
Access	API, LocalSource
Parameters	235BSource
Active parameters	22BSource
Context window	262,144 tokensSource
Weights repository	Qwen/Qwen3-235B-A22B-Thinking-2507Source
License	Apache-2.0Source
Architecture	Qwen3 causal language model with sparse MoE and 8 activated expertsSource
Input	TextSource
Output	TextSource
Reasoning mode	thinking modeSource
Tool calling	YesSource
Structured output	YesSource
Native context window	262,144 tokensSource
Maximum context window	1,010,000 tokensSource
Parameters	235 BSource
Active parameters	22 BSource
Languages	119 languages and dialectsSource
Repository created	2025-07-25 (Repository creation date is a precise publication proxy, but may differ from the public announcement date or an individual checkpoint's first availability.)Source
Training tokens	36 trillionSource

Published benchmarks

SimpleQA Verified v2

Kaggle score

1,000 verified short-answer prompts test factual knowledge without search or other tools. It does not measure document grounding, web research, or a RAG system.

1.Gemini 3.1 Pro Preview77.5%

2.Gemini 3.5 Flash70.4%

3.GPT-5.6 Sol69.2%

4.Claude Opus 563.3%

5.Grok 4.553.8%

6.Qwen 3 235B A22B Thinking 250753.7%

7.o352.3%

higher is betterAs of August 9, 2026Scale: 0% - 100%

Source: Kaggle

CC BY 4.0

gradually.ai

Methodology and context

Comparison cohort: simpleqa-verified-v2:official:overall
Metric: Kaggle leaderboard score
Scale: 0-100
Participants: 7
Methodology and context: 1,000 verified prompts without tools. Kaggle independently reproduced the results.. Source model ID: curated/qwen-3-235b-a22b-thinking. Source model name: Qwen 3 235B A22B Thinking 2507. Provider reported by source: Alibaba. Evaluator: Kaggle. Dataset type: public. Leaderboard date: 2026-08-03. Source ID: google-deepmind. Evidence capture ID: curated-google-deepmind-2026-08-09-47f6b8235215. Source payload hash: 47f6b823521578b4bb8145cf6a2b1768a86464453e8fa38842c288d22daf0f96. Parser version: curated-v1. Imported at: 2026-09-29

Source: Kaggle

Measurements without matching peer values

These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.

MMLU-Pro (Show measurement, test conditions, and source)

Source value: 84.4
Score: 84.4
Metric: percent
Unit: %
Direction: Higher is better
Source type: vendor-reported
Evaluator: Qwen official Hugging Face model card
Status: active
Retrieved at: 2026-09-08
Methodology: Vendor-reported result. Official model-card table; shots/tools/decoding not stated in the card row

Source Qwen official Hugging Face model card

MMLU-Redux (Show measurement, test conditions, and source)

Source value: 93.8
Score: 93.8
Metric: percent
Unit: %
Direction: Higher is better
Source type: vendor-reported
Evaluator: Qwen official Hugging Face model card
Status: active
Retrieved at: 2026-09-08
Methodology: Vendor-reported result. Official model-card table; shots/tools/decoding not stated in the card row

Source Qwen official Hugging Face model card

GPQA (Show measurement, test conditions, and source)

Source value: 81.1
Score: 81.1
Metric: percent
Unit: %
Direction: Higher is better
Source type: vendor-reported
Evaluator: Qwen official Hugging Face model card
Status: active
Retrieved at: 2026-09-08
Methodology: Vendor-reported result. Official model-card table; shots/tools/decoding not stated in the card row

Source Qwen official Hugging Face model card

AIME25 (Show measurement, test conditions, and source)

Source value: 92.3
Score: 92.3
Metric: percent
Unit: %
Direction: Higher is better
Source type: vendor-reported
Evaluator: Qwen official Hugging Face model card
Status: active
Retrieved at: 2026-09-08
Methodology: Vendor-reported result. Official model-card table; shots/tools/decoding not stated in the card row

Source Qwen official Hugging Face model card

LiveBench 20241125 (Show measurement, test conditions, and source)

Source value: 78.4
Score: 78.4
Metric: percent
Unit: %
Direction: Higher is better
Source type: vendor-reported
Evaluator: Qwen official Hugging Face model card
Status: active
Retrieved at: 2026-09-08
Methodology: Vendor-reported result. Official model-card table; shots/tools/decoding not stated in the card row

Source Qwen official Hugging Face model card

OJBench (Show measurement, test conditions, and source)

Source value: 32.5
Score: 32.5
Metric: percent
Unit: %
Direction: Higher is better
Source type: vendor-reported
Evaluator: Qwen official Hugging Face model card
Status: active
Retrieved at: 2026-09-08
Methodology: Vendor-reported result. Official model-card table; shots/tools/decoding not stated in the card row

Source Qwen official Hugging Face model card

IFEval (Show measurement, test conditions, and source)

Source value: 87.8
Score: 87.8
Metric: percent
Unit: %
Direction: Higher is better
Source type: vendor-reported
Evaluator: Qwen official Hugging Face model card
Status: active
Retrieved at: 2026-09-08
Methodology: Vendor-reported result. Official model-card table; shots/tools/decoding not stated in the card row

Source Qwen official Hugging Face model card

Arena-Hard v2$ (Show measurement, test conditions, and source)

Source value: 79.7
Score: 79.7
Metric: percent
Unit: %
Direction: Higher is better
Source type: vendor-reported
Evaluator: Qwen official Hugging Face model card
Status: active
Retrieved at: 2026-09-08
Methodology: Vendor-reported result. Official model-card table; shots/tools/decoding not stated in the card row

Source Qwen official Hugging Face model card

Creative Writing v3 (Show measurement, test conditions, and source)

Source value: 86.1
Score: 86.1
Metric: percent
Unit: %
Direction: Higher is better
Source type: vendor-reported
Evaluator: Qwen official Hugging Face model card
Status: active
Retrieved at: 2026-09-08
Methodology: Vendor-reported result. Official model-card table; shots/tools/decoding not stated in the card row

Source Qwen official Hugging Face model card

BFCL-v3 (Show measurement, test conditions, and source)

Source value: 71.9
Score: 71.9
Metric: percent
Unit: %
Direction: Higher is better
Source type: vendor-reported
Evaluator: Qwen official Hugging Face model card
Status: active
Retrieved at: 2026-09-08
Methodology: Vendor-reported result. Official model-card table; shots/tools/decoding not stated in the card row

Source Qwen official Hugging Face model card

TAU1-Retail (Show measurement, test conditions, and source)

Source value: 67.8
Score: 67.8
Metric: percent
Unit: %
Direction: Higher is better
Source type: vendor-reported
Evaluator: Qwen official Hugging Face model card
Status: active
Retrieved at: 2026-09-08
Methodology: Vendor-reported result. Official model-card table; shots/tools/decoding not stated in the card row

Source Qwen official Hugging Face model card

Recorded measurement details

SimpleQA Verified v2: Kaggle score (Show measurement, test conditions, and source)

Source value: 53.7
Score: 53.7
Metric: Kaggle leaderboard score
Unit: %
Benchmark version: Kaggle leaderboard 2026-08-03
Category: knowledge
Task: overall
Task label: Kaggle score
Direction: Higher is better
Comparison cohort: simpleqa-verified-v2:official:overall
Confidence interval: 3.4
Source type: independent-evaluator
Evaluator: Kaggle
Status: active
Retrieved at: 2026-08-09
Methodology: 1,000 verified prompts without tools. Kaggle independently reproduced the results.. Source model ID: curated/qwen-3-235b-a22b-thinking. Source model name: Qwen 3 235B A22B Thinking 2507. Provider reported by source: Alibaba. Evaluator: Kaggle. Dataset type: public. Leaderboard date: 2026-08-03. Source ID: google-deepmind. Evidence capture ID: curated-google-deepmind-2026-08-09-47f6b8235215. Source payload hash: 47f6b823521578b4bb8145cf6a2b1768a86464453e8fa38842c288d22daf0f96. Parser version: curated-v1. Imported at: 2026-09-29
Comparison methodology: 1,000 verified prompts without tools. Kaggle independently reproduced the results.

Source Kaggle

Page 1 of 3

Sources and data date

Type	Evidence and data date
Speed	Alibaba. Output speed after the first streaming chunk, not total response time. Classified as deprecated by Artificial Analysis. Artificial Analysis maintains only the default 10K-input workload there. Retrieved September 8, 2026. Source: Artificial Analysis · Artificial Analysis (retrieved September 8, 2026)Speed measurement methodology
Price	Base API prices without caching or batch discounts. Higher context tiers and other rates are listed under Costs. Source: Pricing · API pricing
Input	Documented modalities. Retrieved September 8, 2026. Source: Model documentation · Model metadata (retrieved September 8, 2026) · Qwen3-235B-A22B Thinking 2507 model card · Editorial description reviewed October 2, 2026
Output	Documented modalities. Retrieved September 8, 2026. Source: Model documentation · Model metadata (retrieved September 8, 2026) · Qwen3-235B-A22B Thinking 2507 model card · Editorial description reviewed October 2, 2026
Research status	Research date September 8, 2026. 4 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication.
Additional source	Kaggle (retrieved August 9, 2026)
Additional source	Model metadata (retrieved September 8, 2026) · Qwen3 Technical Report · Editorial description reviewed October 2, 2026
Additional source	Model metadata (retrieved September 8, 2026)
Additional source	OpenRouter: Popularity index (derived), CC BY 4.0
Additional source	Vercel AI Gateway: Popularity index (derived), CC BY 4.0
Additional source	Creative Commons Attribution 4.0

Qwen 3 235B A22B Thinking 2507

Specifications and access

Published benchmarks

SimpleQA Verified v2

Measurements without matching peer values

Recorded measurement details

Our tests

Pelican on a bicycle

Popularity

Published prices

Sources and data date