DeepSeek V3.2 Exp (Reasoning) vs Gemini 3 Pro Preview (high)

Comparing 2 AI models · 5 benchmarks · DeepSeek, Google

Most Affordable
DeepSeek logo
DeepSeek V3.2 Exp (Reasoning)
$0.28/1M
Highest Intelligence
Google logo
Gemini 3 Pro Preview (high)
90.8% GPQA
Best for Coding
Google logo
Gemini 3 Pro Preview (high)
62.3 Coding Index
Price Difference
7.1x
input cost range

Composite Indices

Intelligence, Coding, Math

Standard Benchmarks

Academic and industry benchmarks

Benchmark Winners

5 tests
DeepSeek logo

DeepSeek V3.2 Exp (Reasoning)

0

No clear wins

Google logo

Gemini 3 Pro Preview (high)

5
  • GPQA
  • MMLU Pro
  • HLE
  • LiveCodeBench
  • AIME 2025
Metric
DeepSeek logo DeepSeek V3.2 Exp (Reasoning)
DeepSeek
Google logo Gemini 3 Pro Preview (high)
Google
Pricing
Per 1M tokens
Input Cost $0.28/1M $2.00/1M
Output Cost $0.42/1M $12.00/1M
Blended Cost 3:1 input/output ratio
$0.32/1M $4.50/1M
Specifications
Organization Model creator
DeepSeek Google
Release Date Launch date
Sep 29, 2025 Nov 18, 2025
Performance & Speed
Throughput Output speed
29.1 tok/s 138.9 tok/s
Time to First Token (TTFT) Initial response delay
1249ms 26684ms
Latency Time to first answer token
69976ms 26684ms
Composite Indices
Intelligence Index Overall reasoning capability
56.9 72.8
Coding Index Programming ability
48.6 62.3
Math Index Mathematical reasoning
87.7 95.7
Standard Benchmarks
GPQA Graduate-level reasoning
79.7% 90.8%
MMLU Pro Advanced knowledge
85.0% 89.8%
HLE Hard language evaluation
13.8% 37.2%
LiveCodeBench Real-world coding tasks
78.9% 91.7%
MATH 500 Mathematical problems
AIME 2025 Advanced math competition
87.7% 95.7%
AIME (Original) Math olympiad problems
SciCode Scientific code generation
37.7% 56.1%
LCR Code review capability
69.0% 70.7%
IFBench Instruction-following
54.1% 70.4%
TAU-bench v2 Tool use & agentic tasks
33.9% 87.1%
TerminalBench Hard CLI command generation
29.1% 39.0%

Key Takeaways

DeepSeek V3.2 Exp (Reasoning) offers the best value at $0.28/1M, making it ideal for high-volume applications and cost-conscious projects.

Gemini 3 Pro Preview (high) leads in reasoning capabilities with a 90.8% GPQA score, excelling at complex analytical tasks and problem-solving.

Gemini 3 Pro Preview (high) achieves a