modelgrep
Q

Qwen: Qwen3 30B A3B Thinking 2507

qwen/qwen3-30b-a3b-thinking-2507

111th smartest of 180Cheaper than 87% of paidReasoningToolsJSON
Use via OpenRouter ↗
Intelligence
22.4
111th of 180
Design Elo
975
Website
Speed
134
34th fastest
Latency
499ms
first token
Input price
$0.080
39th cheapest
Context
131K
131K max out

How it compares

Smarter than38%
of all ranked models
Faster than89%
of all ranked models
Cheaper than87%
of all ranked models

Overview

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...

Benchmarks

independent · via OpenRouter
Artificial Analysis37th percentile
Intelligence Index
22.4
Coding Index
14.6
Agentic Index
17.7
GPQA Diamond
71%
Humanity's Last Exam
10%
SciCode
33%
Tau²-Bench (agentic)
28%
Design Arena · Elo377 tournaments
Website
975
Data Viz
969

Providers & pricing (2)

ProviderIn $/MOut $/MUptime
AtlasCloudfp8$0.080$0.400100%
Alibaba$0.130$1.56

Specifications

Context window131K
Max output131K
Knowledge cutoffJun 2025
Input modalitiestext
Output modalitiestext
Prompt caching
Cache read price$0.080/M
ModeratedNo

Qwen3 30B A3B Thinking 2507 FAQ

How much does Qwen3 30B A3B Thinking 2507 cost?

Qwen3 30B A3B Thinking 2507 costs $0.080 per million input tokens and $0.400 per million output tokens via OpenRouter, making it 39th cheapest of 298 paid models.

How smart is Qwen3 30B A3B Thinking 2507?

Qwen3 30B A3B Thinking 2507 scores 22.4 on the Artificial Analysis Intelligence Index, ranking 111th of 180 benchmarked models, with a GPQA Diamond score of 71%.

How fast is Qwen3 30B A3B Thinking 2507?

Qwen3 30B A3B Thinking 2507 generates around 134 tokens per second with 499ms time-to-first-token (p50), the 34th fastest tracked model.

What is Qwen3 30B A3B Thinking 2507's context window?

Qwen3 30B A3B Thinking 2507 supports a 131K-token context window and can output up to 131K tokens. It accepts text input.

Compare head-to-head