BenchAtlas

Rankings / DeepSeek

Deepseek V3 1 Terminus

open weights

released 2025-09-22

BenchAtlas Index

as of 2026-09-11
45.7
thinking
rank #270
9 families · 5 categories · high
35.9
no reasoning
rank #334
9 families · 5 categories · high

Benchmark evidence

39 results
agentic coding
τ²-Benchthinking37.1 %independentArtificial Analysis
τ²-Benchno reasoning37.1 %independentArtificial Analysis
τ²-Bench
subset=banking
thinking21.0 %independentArtificial Analysis
coding
LiveCodeBench v6
implementation=artificial-analysis
thinking79.8 %independentArtificial Analysis
LiveCodeBench v6
implementation=artificial-analysis
no reasoning52.9 %independentArtificial Analysis
SciCodethinking38.0 %independentArtificial Analysis
SciCodeno reasoning32.1 %independentArtificial Analysis
external indices
AA Coding Indexthinking43.5 pointsindependentArtificial Analysis
Intelligence Index v4.1thinking15.4 pointsindependentArtificial Analysis
Intelligence Index v4.1no reasoning13.9 pointsindependentArtificial Analysis
AA Math Indexthinking89.7 pointsindependentArtificial Analysis
AA Math Indexno reasoning53.7 pointsindependentArtificial Analysis
human preference
Industry Life And Physical And Social Science (style control)older version
arena=text · category=industry_life_and_physical_and_social_science · style_control=true
1452.9
1428.6–1477.2
communityLMArena Leaderboard Dataset
Coding (style control)older version
arena=text · category=coding · style_control=true
1440.1
1419.5–1460.7
communityLMArena Leaderboard Dataset
Industry Software And It Services (style control)older version
arena=text · category=industry_software_and_it_services · style_control=true
1428.2
1412.1–1444.3
communityLMArena Leaderboard Dataset
Hard Prompts English (style control)older version
arena=text · category=hard_prompts_english · style_control=true
1424.7
1405.0–1444.5
communityLMArena Leaderboard Dataset
Hard Prompts (style control)older version
arena=text · category=hard_prompts · style_control=true
1423.2
1410.0–1436.4
communityLMArena Leaderboard Dataset
Longer Query (style control)older version
arena=text · category=longer_query · style_control=true
1418.8
1399.1–1438.5
communityLMArena Leaderboard Dataset
English (style control)older version
arena=text · category=english · style_control=true
1418.4
1403.7–1433.0
communityLMArena Leaderboard Dataset
Overall (style control)
arena=text · category=overall · style_control=true
1414.7
1405.0–1424.4
communityLMArena Leaderboard Dataset
Industry Writing And Literature And Language (style control)older version
arena=text · category=industry_writing_and_literature_and_language · style_control=true
1406.8
1386.0–1427.7
communityLMArena Leaderboard Dataset
Non English (style control)older version
arena=text · category=non_english · style_control=true
1403.4
1390.6–1416.2
communityLMArena Leaderboard Dataset
Exclude Ties (style control)older version
arena=text · category=exclude_ties · style_control=true
1399.8
1385.9–1413.8
communityLMArena Leaderboard Dataset
Industry Business And Management And Financial Operations (style control)older version
arena=text · category=industry_business_and_management_and_financial_operations · style_control=true
1395.8
1374.2–1417.5
communityLMArena Leaderboard Dataset
Multi Turn (style control)older version
arena=text · category=multi_turn · style_control=true
1395.2
1372.6–1417.7
communityLMArena Leaderboard Dataset
Instruction Following (style control)older version
arena=text · category=instruction_following · style_control=true
1393.2
1375.4–1411.1
communityLMArena Leaderboard Dataset
Industry Entertainment And Sports And Media (style control)older version
arena=text · category=industry_entertainment_and_sports_and_media · style_control=true
1376.6
1353.5–1399.8
communityLMArena Leaderboard Dataset
knowledge science
GPQA Diamond
implementation=artificial-analysis
thinking79.2 %independentArtificial Analysis
GPQA Diamond
implementation=artificial-analysis
no reasoning75.1 %independentArtificial Analysis
Humanity's Last Exam
implementation=artificial-analysis
thinking16.4 %independentArtificial Analysis
Humanity's Last Exam
implementation=artificial-analysis
no reasoning8.7 %independentArtificial Analysis
MMLU-Pro
implementation=artificial-analysis
thinking85.1 %independentArtificial Analysis
MMLU-Pro
implementation=artificial-analysis
no reasoning83.6 %independentArtificial Analysis
long context instruction
AA-LCRthinking69.3 %independentArtificial Analysis
AA-LCRno reasoning45.3 %independentArtificial Analysis
IFBenchthinking57.0 %independentArtificial Analysis
IFBenchno reasoning41.2 %independentArtificial Analysis
reasoning math
AIME
year=2025 · implementation=artificial-analysis
thinking89.7 %independentArtificial Analysis
AIME
year=2025 · implementation=artificial-analysis
no reasoning53.7 %independentArtificial Analysis

Agent + model results

systems, not bare-model scores
agent + model Artificial Analysis harness + Deepseek V3 1 TerminusTerminal-Bench 2.144.9 %independentArtificial Analysis
agent + model Artificial Analysis harness + Deepseek V3 1 TerminusTerminal-Bench Hard31.8 %independentArtificial Analysis
agent + model Artificial Analysis harness + Deepseek V3 1 TerminusTerminal-Bench Hard30.3 %independentArtificial Analysis

These scores measure the whole agent system (scaffold, tools, budgets) — they are never merged into the bare model’s numbers.