This website uses cookies for authentication, support chat, and analytics. Only authentication cookies are turned on by default.
Animate this data
Bar chart race of DeepSeek, Google DeepMind, OpenAI and 7 more, by GPQA Diamond.
Create it free
Takes a free account
DeepSeek
DeepSeek has risen from 24.6% in 2023 to 93.2% in 2026.
Aider Polyglot
74 %
Ale bench
1,622
Algotune
2 %
Apex Agents
79 %
Arc Agi
95 %
Arc Agi 2
78 %
BALROG
35 %
Bbh
88 %
Chess Puzzles
48 %
Cl bench
13 %
Cl bench Life
14 %
Critpt
24 %
CursorBench
48 %
DeepResearchBench
35 %
Dtbench
94 %
Forecastbench
63 %
FrontierCode
44 %
FrontierMath
22 %
FrontierMath Tier 4
2 %
FrontierMath Tier 4 V2
46 %
FrontierMath Tiers 1 3 V2
71 %
Furniture Assembly
34 %
GPQA Diamond
93 %
Gsm8k
95 %
Hella Swag
89 %
Lech Mazur Writing
8 %
Live bench
72 %
Lmca
47 %
Math Level 5
97 %
Mmlu
87 %
Mystery Game Puzzles
49 %
OTIS Mock AIME 2024 2025
99 %
Piqa
85 %
Proofbench
56 %
SciCode
56 %
SimpleBench
61 %
SimpleQA Verified
53 %
Surface Evolver bench
46 %
SWE bench Verified
78 %
TerminalBench
40 %
Trivia Qa
83 %
Vending bench 2
3,285
WebDev Arena
1,614
WeirdML
66 %
WeirdML V3
9 %
Wino Grande
86 %
Training Compute
9,702,000,000,000.00T FLOP
DeepSeek benchmark results over time.