Benchmark Almanac

DeepSeek-V3.1-Terminus

DeepSeek-V3.1-Terminus places in the top three on 3 benchmarks.

DeepSeek Released September 22nd, 2025 Family DeepSeek-V3 The lab's model card Compare with others
28
Benchmarks with a score
0
Ranked first, out of five or more models
3
In the top three
DeepSeek Sep 2025

DeepSeek-V3.1-Terminus

Open weights Flagship

144.1 #140 of 404
Self-Reported Index, from the lab's own numbers. No Epoch Capabilities Index score yet.
  • BrowseComp-ZH 45% #19 /19
  • Meta Agents Research Environments (ARE) / Gaia2 23.1% #4 /13
  • MCP-Universe 21.6% #22 /30

Where it ranks

Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.

Knowledge and language

BenchmarkScoreRankSource
SimpleQA 96.8% #2 of 60 Lab model card
Encyclo-K 49.8% #3 of 23 Official leaderboard
MMLU-Pro 85% #28 of 167 Lab model card
LMArena (Elo) 1,416 #110 of 335 Official leaderboard
Humanity's Last Exam 21.7% #101 of 154 Lab model card

Math and science

BenchmarkScoreRankSource
GPQA Diamond 80.7% #115 of 302 Lab model card
DTBench 81.3% #85 of 180 Official leaderboard

Coding

BenchmarkScoreRankSource
Aider Polyglot 76.1% #6 of 30 Lab model card
BigCodeArena 1,076 #3 of 13 Official leaderboard
LiveCodeBench 74.9% #13 of 54 Lab model card
StructEval 77.2% #7 of 28 Official leaderboard
Codeforces rating 2,046 #24 of 47 Lab model card
SWE-bench Verified 68.4% #89 of 155 Lab model card
CC-Bench 27% #3 of 4 Official leaderboard
ALE-Bench 1,078 #97 of 126 Official leaderboard
SWE-bench Multilingual 57.8% #62 of 73 Lab model card

Agents and tools

BenchmarkScoreRankSource
Meta Agents Research Environments (ARE) / Gaia2 23.1% #4 of 13 Official leaderboard
MCPMark 21.3% #15 of 28 Official leaderboard
SnakeBench 21 #122 of 188 Official leaderboard
CompileBench 73.3% #14 of 20 Official leaderboard
Terminal-Bench 36.7% #16 of 22 Lab model card
MCP-Universe 21.6% #22 of 30 Official leaderboard
BrowseComp 38.5% #73 of 81 Lab model card
BrowseComp-ZH 45% #19 of 19 Lab model card

Safety and security

BenchmarkScoreRankSource
nocot-bench (NCRI) 77.6 #64 of 185 Official leaderboard

Other

BenchmarkScoreRankSource
Conceptual Reasoning Index 42.0 #95 of 151 Official leaderboard
LMCA 28.6% #98 of 147 Official leaderboard
ACCoRD 55.2 #107 of 152 Official leaderboard

Its lab's numbers put it at 144.1, number 140 of 404

The Self-Reported Index fits one score to the 11 capability results DeepSeek reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 139.7 to 148.5. How we compute it.

Its launch post

The benchmarks the lab chose to headline, with the scores it gave.