ModelsCompareBest forBenchmarksStatusPricingAPI

DeepSeek V3.2

DeepSeek·Operational·Released 2025-12-01text
Context131.072k
Max output65.536k
Input / 1M-
Output / 1M-
Cached in-
Benchmarks8

Benchmarks

Cited public sources · rank is exact-benchmark, apples-to-apples

AIME 2025 (MathArena)#5/2394.2% ↗
AIME 2026 (MathArena)#17/2194.2% ↗
LMArena Elo#34/621424.8 ↗
AIME 2025#3/1593.1% ↗
GPQA Diamond#42/7082.4% ↗
Humanity's Last Exam (no tools)#17/2925.1% ↗
MMLU-Pro#8/2985% ↗
SWE-bench Verified#23/4573.1% ↗

Compare

vs GPT-6 Astravs GPT-5.6 Solvs GPT-5.6 Terravs GPT-5.6 Lunavs Claude Fable 5.1vs Claude Opus 5vs Claude Sonnet 5vs Claude Haiku 4.5

More from DeepSeek

DeepSeek V4.1 FlashDeepSeek V4 ProDeepSeek R1DeepSeek V3

Access DeepSeek V3.2 and every other model through one endpoint with automatic failover: Respan gateway.