Qwen: Qwen3 30B A3B Thinking 2507

Text input Text output
Author's Description

Qwen3-30B-A3B-Thinking-2507 is a 30B parameter Mixture-of-Experts reasoning model optimized for complex tasks requiring extended multi-step thinking. The model is designed specifically for “thinking mode,” where internal reasoning traces are separated from final answers. Compared to earlier Qwen3-30B releases, this version improves performance across logical reasoning, mathematics, science, coding, and multilingual benchmarks. It also demonstrates stronger instruction following, tool use, and alignment with human preferences. With higher reasoning efficiency and extended output budgets, it is best suited for advanced research, competitive problem solving, and agentic applications requiring structured long-context reasoning.

Key Specifications
Cost
$$$$
Context
262K
Parameters
30B
Released
Aug 28, 2025
Speed
Ability
Reliability
Supported Parameters

This model supports the following parameters:

Top Logprobs Stop Max Tokens Tool Choice Top P Frequency Penalty Reasoning Seed Include Reasoning Logprobs Logit Bias Tools Temperature Presence Penalty
Features

This model supports the following features:

Reasoning Tools
Performance Summary

Qwen3-30B-A3B-Thinking-2507 demonstrates exceptional performance across several key metrics. It consistently ranks among the fastest models available and offers highly competitive pricing, making it an efficient and cost-effective solution. The model exhibits outstanding reliability with a 100% success rate across all benchmarks, indicating minimal technical failures. In terms of specific performance, Qwen3-30B-A3B-Thinking-2507 excels in complex reasoning tasks, achieving a 98.0% accuracy (95th percentile) in Reasoning and a perfect 100.0% in Ethics, where it is noted as the most accurate model at its price point and among models of similar speed. It also shows strong capabilities in Coding (92.9% accuracy, 83rd percentile), Mathematics (93.9% accuracy, 84th percentile), and General Knowledge (99.5% accuracy, 79th percentile). A notable weakness is its Instruction Following, where it scored 0.0% accuracy, suggesting a significant area for improvement. Its hallucination rate, while not the lowest at 84.0% accuracy, indicates room for improvement in acknowledging uncertainty. Overall, its strengths lie in advanced problem-solving and ethical alignment, making it well-suited for research and agentic applications.

Model Pricing

Current Pricing

Feature Price (per 1M tokens)
Prompt $0.1
Completion $0.3

Price History

Available Endpoints
Provider Endpoint Name Context Length Pricing (Input) Pricing (Output)
Nebius
Nebius | qwen/qwen3-30b-a3b-thinking-2507 262K $0.1 / 1M tokens $0.3 / 1M tokens
Chutes
Chutes | qwen/qwen3-30b-a3b-thinking-2507 262K $0.08 / 1M tokens $0.29 / 1M tokens
Alibaba
Alibaba | qwen/qwen3-30b-a3b-thinking-2507 131K $0.2 / 1M tokens $2.4 / 1M tokens
SiliconFlow
SiliconFlow | qwen/qwen3-30b-a3b-thinking-2507 262K $0.09 / 1M tokens $0.3 / 1M tokens
Chutes
Chutes | qwen/qwen3-30b-a3b-thinking-2507 262K $0.08 / 1M tokens $0.29 / 1M tokens
Benchmark Results
Benchmark Category Reasoning Strategy Free Executions Accuracy Cost Duration
Other Models by qwen