DeepSeek V3.2 vs DeepSeek V4

Benchmarks, API pricing and specs, head to head. Data updated 2026-06-10.

DeepSeek V3.2

DeepSeek · Sep 2025

68.4

The sparse-attention release that halved DeepSeek's already-lowest-in-class prices while keeping GPT-class quality.

DeepSeek V4

DeepSeek · Mar 2026

The open-weights shock of 2026: ~81% SWE-bench Verified, a 1M context and MIT license at $0.30/$0.50 per million tokens.

The verdict

DeepSeek V4 wins 1 of the 1 benchmarks these models share, against 0 for DeepSeek V3.2. DeepSeek V3.2 is about 1.1x cheaper per blended million tokens (3:1 input:output mix). DeepSeek V4 also takes 1M of context versus 128K for DeepSeek V3.2.

Benchmark head-to-head 01

SWE-bench Verified
67.8%~81%
DeepSeek V3.2DeepSeek V4

Specs & pricing

DeepSeek V3.2DeepSeek V4
modhub Index68.4
Input price / 1M$0.28$0.3
Output price / 1M$0.42$0.5
Context window128K1M
Max output64K64K
Open weightsyes (MIT)yes (MIT)
Reasoning modelyesyes
Multimodal inputtexttext
Knowledge cutoffJul 2025Dec 2025
ReleasedSep 2025Mar 2026
Example monthly cost*$3.43$3.75

* 10M input + 1.5M output tokens per month at list prices, no caching. Green = better value on that row.

Frequently asked questions

Which is better, DeepSeek V3.2 or DeepSeek V4?
DeepSeek V4 wins 1 of the 1 benchmarks these models share, against 0 for DeepSeek V3.2. DeepSeek V3.2 is about 1.1x cheaper per blended million tokens (3:1 input:output mix). DeepSeek V4 also takes 1M of context versus 128K for DeepSeek V3.2.
Which is cheaper, DeepSeek V3.2 or DeepSeek V4?
DeepSeek V3.2 costs $0.28/$0.42 per million input/output tokens, while DeepSeek V4 costs $0.3/$0.5. For a typical workload of 10M input and 1.5M output tokens per month, that's $3.43 versus $3.75.
Which model is better for coding, DeepSeek V3.2 or DeepSeek V4?
On SWE-bench Verified — the standard agentic-coding benchmark — DeepSeek V4 scores ~81% versus 67.8% for DeepSeek V3.2, making DeepSeek V4 the stronger pick for coding agents.

More comparisons