DeepSeek-V3.2 (Thinking) vs DeepSeek-V4-Pro-0813: Specs & Benchmark Comparison

DeepSeek-V3.2 (Thinking) and DeepSeek-V4-Pro-0813 both come from DeepSeek, DeepSeek-V3.2 (Thinking) was released in November 2025, and DeepSeek-V4-Pro-0813 followed 9 months later in August 2026. DeepSeek-V4-Pro-0813 is the larger model at roughly 1.6 trillion parameters, against 685 billion for DeepSeek-V3.2 (Thinking).

The two models share 2 published benchmarks. DeepSeek-V4-Pro-0813 leads on 2 of them. The widest gaps are on Toolathlon, where DeepSeek-V4-Pro-0813 scores 74.0% against 35.2%; Humanity's Last Exam, where DeepSeek-V4-Pro-0813 scores 60.0% against 25.1%. Averaged across everything we track, DeepSeek-V3.2 (Thinking) sits at 68.5% and DeepSeek-V4-Pro-0813 at 60.6%.

DeepSeek-V3.2 (Thinking) is the cheaper API at $0.28 per million input tokens and $0.42 per million output tokens, roughly 5 times cheaper than DeepSeek-V4-Pro-0813 at $1.32 and $3.96. DeepSeek-V4-Pro-0813 takes the larger context window at 1M tokens, compared with 131K for DeepSeek-V3.2 (Thinking).

CharacteristicDeepSeek-V3.2 (Thinking)DeepSeek-V4-Pro-0813
CompanyDeepSeekDeepSeek
Release DateNovember 30, 2025August 13, 2026
Parameters685B1.6T
MultimodalNoNo
Context (input)131K1.0M
Context (output)66K393K
Input Price / 1M$0.28$1.32
Output Price / 1M$0.42$3.96
Average Score68.5%60.6%
Benchmarks
Toolathlon35.2%74.0%
Humanity's Last Exam25.1%60.0%

Visual Benchmark Comparison

DeepSeek-V3.2 (Thinking)
DeepSeek-V4-Pro-0813
Toolathlon0.4 vs 0.7
0.4
0.7
Humanity's Last Exam0.3 vs 0.6
0.3
0.6

Verdict

DeepSeek-V4-Pro-0813 leads in 2 out of 4 comparison categories.

Overall Performance

Both models show comparable average scores: DeepSeek-V3.2 (Thinking) — 0.7, DeepSeek-V4-Pro-0813 — 0.6.

API Cost

DeepSeek-V3.2 (Thinking) is 7.5x cheaper: input $0.28/1M vs $1.32/1M tokens.

Context Window

DeepSeek-V4-Pro-0813 supports a larger context: 1M vs 131K tokens.

Recency

DeepSeek-V4-Pro-0813 is newer: released 8/13/2026 vs 11/30/2025.

More About These Models

Related Comparisons

Frequently Asked Questions

Which is better for coding — DeepSeek-V3.2 (Thinking) or DeepSeek-V4-Pro-0813?
Direct comparison on the SWE-Bench benchmark is not available. We recommend reviewing other metrics on the comparison page.
Which model is cheaper — DeepSeek-V3.2 (Thinking) or DeepSeek-V4-Pro-0813?
DeepSeek-V3.2 (Thinking) is cheaper for input: $0.28 per 1M tokens vs $1.32.
Which has a larger context window — DeepSeek-V3.2 (Thinking) or DeepSeek-V4-Pro-0813?
DeepSeek-V4-Pro-0813 supports a larger context: 1,048,576 tokens vs 131,072.

The DeepSeek-V3.2 (Thinking) and DeepSeek-V4-Pro-0813 comparison is updated for 2026. Data includes benchmark results, API pricing, context window size and other specifications. For more detailed information, visit the DeepSeek-V3.2 (Thinking) or DeepSeek-V4-Pro-0813 page. See also the complete list of AI model comparisons.