deepseek-v3.2-exp-thinking vs deepseek-v4-pro Benchmark Comparison

Direct benchmark comparison between deepseek-v3.2-exp-thinking and deepseek-v4-pro based on LMArena Elo and the latest 2026 API pricing.

Direct Technical & Pricing Comparison

Frontier Model	LMArena Elo	API Cost (1M)	Throughput
deepseek-v3.2-exp-thinking	1425	$2.7e-7	8
deepseek-v4-pro	1463	$0.00000174	25

*These models represent the Pareto Frontier (optimal cost-to-performance).*

Comparison Summary: deepseek-v4-pro is the more capable model in this pair, leading by 38 Elo points. However, neither model is currently Pareto-optimal. Developers looking for peak efficiency should investigate claude-opus-4-7-thinking, which offers a superior benchmark-to-price ratio.