Qwen3-235B-A22B-Thinking-2507 is a high-performance, open-weight Mixture-of-Experts (MoE) language model
optimized for complex reasoning tasks. It activates 22B of its 235B parameters per forward pass and natively
supports up to 262,144 tokens of context. This "thinking-only" variant enhances structured logical reasoning,
mathematics, science, and long-form generation, showing strong benchmark performance across AIME, SuperGPQA,
LiveCodeBench, and MMLU-Redux. It enforces a special reasoning mode (</think>) and is designed for high-token
outputs (up to 81,920 tokens) in challenging domains.
The model is instruction-tuned and excels at step-by-step reasoning, tool use, agentic workflows, and multilingual
tasks. This release represents the most capable open-source variant in the Qwen3-235B series, surpassing many closed
models in structured reasoning use cases
Evaluations
10
across 9 benchmarks
Latency
2m 30s
Context length
262k
tokens
Cost
$0.23 · $2.30
input · output per 1M tokens
Key takeaways
Qwen3 235B A22B Thinking 2507 is a high-performance, open-weight Mixture-of-Experts (MoE) Transformer language model optimized for complex reasoning tasks, specifically for structured logical reasoning, mathematics, science, and long-form generation.
The model is instruction-tuned and supports context lengths up to 262,144 tokens with high-token outputs (up to 81,920 tokens), excelling at step-by-step reasoning, tool use, agentic workflows, and multilingual tasks.
It demonstrated strong foundational logical reasoning and achieved a perfect 100% accuracy on the AIME 2024 benchmark, although this was accompanied by a technical issue where the model's output results were not captured.
A critical limitation is its complete failure (0% success rate) and absence of output for all 28 problems on the AIME 2025 benchmark, indicating a significant issue in task execution or processing complex mathematical reasoning.
The model struggles with nuanced causal judgment, particularly preemption, distinguishing fine-grained error types in translation, and differentiating between intentional actions and unintended side effects, suggesting challenges in deeper semantic understanding.
