The Model Index
Up to four models

DeepSeek-V3.1 vs DeepSeek-V3 (Mar 2025)

On the index (GPT-4 = 100), DeepSeek-V3.1 scores 128 and DeepSeek-V3 (Mar 2025) 124. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.

Next to DeepSeek-V3.1
Presets
DeepSeek-V3.1
128111–145
WeightsOpen
ReleasedAug 2025
Price in / out$0.25 / $0.95
Context164K
Core tests taken0 of 9
DeepSeek-V3 (Mar 2025)
124108–140
WeightsOpen
ReleasedMar 2025
Price in / out$0.29 / $1.14
Context164K
Core tests taken2 of 9

DeepSeek-V3.1 and DeepSeek-V3 (Mar 2025) are too close to call (4 points apart, within the uncertainty), at a similar price.

Test by testthird-party results

DeepSeek-V3.1DeepSeek-V3 (Mar 2025)
0%25%50%75%100%REASONING & KNOWLEDGEFiction.LiveBenchFiction.LiveBenchSimpleBenchSimpleBenchCODINGWeirdMLWeirdMLAGENTSDTBenchDTBenchLMCALMCA
Every result as a table
TestDeepSeek-V3.1DeepSeek-V3 (Mar 2025)
Reasoning & knowledge
Fiction.LiveBench52.8%50.0%
SimpleBench40.0%27.2%
Coding
WeirdML38.4%36.1%
Agents
DTBench82.7%64.8%
LMCA24.3%15.5%
Task length · METR30 sec

Tests at least two of these models have taken. Bold rows are the core tests.

This page as a chartDownload PNG