Qwen3 235B A22B 2507 Instruct vs Qwen3 235B A22B 2507 (Reasoning)

Comparing 2 AI models · 6 benchmarks · Alibaba

Most Affordable
Alibaba logo
Qwen3 235B A22B 2507 Instruct
$0.70/1M
Highest Intelligence
Alibaba logo
Qwen3 235B A22B 2507 (Reasoning)
79.0% GPQA
Best for Coding
Alibaba logo
Qwen3 235B A22B 2507 (Reasoning)
44.6 Coding Index
Price Difference
1.0x
input cost range

Composite Indices

Intelligence, Coding, Math

Standard Benchmarks

Academic and industry benchmarks

Benchmark Winners

6 tests
Alibaba logo

Qwen3 235B A22B 2507 Instruct

0

No clear wins

Alibaba logo

Qwen3 235B A22B 2507 (Reasoning)

6
  • GPQA
  • MMLU Pro
  • HLE
  • LiveCodeBench
  • MATH 500
  • AIME 2025
Metric
Alibaba logo Qwen3 235B A22B 2507 Instruct
Alibaba
Alibaba logo Qwen3 235B A22B 2507 (Reasoning)
Alibaba
Pricing
Per 1M tokens
Input Cost $0.70/1M $0.70/1M
Output Cost $2.80/1M $8.40/1M
Blended Cost 3:1 input/output ratio
$1.22/1M $2.63/1M
Specifications
Organization Model creator
Alibaba Alibaba
Release Date Launch date
Jul 21, 2025 Jul 25, 2025
Performance & Speed
Throughput Output speed
39.3 tok/s 82.6 tok/s
Time to First Token (TTFT) Initial response delay
1254ms 1136ms
Latency Time to first answer token
1254ms 25354ms
Composite Indices
Intelligence Index Overall reasoning capability
45.3 57.5
Coding Index Programming ability
34.2 44.6
Math Index Mathematical reasoning
71.7 91.0
Standard Benchmarks
GPQA Graduate-level reasoning
75.3% 79.0%
MMLU Pro Advanced knowledge
82.8% 84.3%
HLE Hard language evaluation
10.6% 15.0%
LiveCodeBench Real-world coding tasks
52.4% 78.8%
MATH 500 Mathematical problems
98.0% 98.4%
AIME 2025 Advanced math competition
71.7% 91.0%
AIME (Original) Math olympiad problems
71.7% 94.0%
SciCode Scientific code generation
36.0% 42.4%
LCR Code review capability
31.2% 67.0%
IFBench Instruction-following
46.1% 51.2%
TAU-bench v2 Tool use & agentic tasks
33.3% 53.2%
TerminalBench Hard CLI command generation
14.2% 12.8%

Key Takeaways

Qwen3 235B A22B 2507 Instruct offers the best value at $0.70/1M, making it ideal for high-volume applications and cost-conscious projects.

Qwen3 235B A22B 2507 (Reasoning) leads in reasoning capabilities with a