DeepSeek V3.2 vs DeepSeek V4.1-Flash

Best cell per row highlighted. Null means undisclosed — never counted as zero.

DeepSeek
DeepSeek V3.2GA
8.2
AI Panel
DeepSeek
DeepSeek V4.1-FlashGA
8.6
AI Panel

Identity & lifecycle

ProviderDeepSeekDeepSeek
Family / tierDeepSeek V3 · LargeDeepSeek V4 · Flash
StatusGAGA
Released2025-12-012026-09-10
Knowledge cutoff2025-07—

Architecture & context

Context window128K1M
Max output tokens64K384K
Input modalitiestexttext, image
Reasoning modeoptionaloptional
Open weightsYesYes

Pricing (per Mtok)

Input$0.252$0.3
Output$0.378$1.2
Cached input$0.025$0.006
Batch input——
Free tierNoYes

Speed

Output speed (tok/s)——
Time to first token (s)——
Latency tiermediumfast

Trust & deployment

Trains on inputsYesYes
LicenseMITMIT
Clouds——
Compliance——

AI Panel scoring

Unified score8.28.6
Decision Maker88.5
Domain Strategist89
Finance Lead99
Domain Practitioner8.59
Power User7.58.5
Skeptic7.57.5
Value score9.39.6

Benchmarks

MMLU-Pro85—
GPQA Diamond—90.9
AIME 202593.1—
SWE-bench Verified67.8—
LiveCodeBench74.1—
Terminal-Bench—90.6
Humanity's Last Exam30.636.8
LMArena Elo1425—
LMArena Coding Elo1476—

Scores link to their sources. Missing cells mean the vendor hasn't disclosed a result — honesty over padding.