DeepSeek-V3

DeepSeekReleased 26 Dec 2024Open

Compare Announcement

116105–128Index score#145 of 229 models

2/ 8Core tests taken2 of 6 areas · plus 6 anchor tests

−20Versus the frontier at releasebest then: o1Replaced by DeepSeek-V3 (Mar 2025)

$0.26/ $1.03Price per million tokensinput / output, list price

164KContext windowtokens it reads at once

Where it sitson the frontier

o1-mini reached this score 3 months earlier.

Test resultsby area · tick = best by any model

Reasoning & knowledgeevidence

GPQA Diamond

57%

Humanity's Last Exam

no result

Mathsevidence

FrontierMath T1-3

no result

Codingevidence

SWE-bench Verified

no result

Terminal-Bench

no result

Agentsno evidence yet

APEX-Agents

no result

Novel problemsno evidence yet

ARC-AGI-2

no result

Long tasksevidence

METR time horizon

18 min

Independent results from the Epoch AI benchmarking hub, best reasoning setting. The core tests are shown here; anchor tests also inform the score and can be shown above.

What others sayrank on each leaderboard

The Model Indexindependent tests, our method#145of 219 · 116

LMArena · Creative writingcreative writing#159of 386 · 1349 LMArena · Textoverall chat quality, judged by people#183of 388 · 1358 Epoch Capabilities Indexoverall capability from many benchmarks#118of 239 · 135.9 LMArena · Codingcoding questions in chat#200of 383 · 1387 LMArena · Mathsmaths questions in chat#213of 373 · 1310

Further left is better. Leaderboards measure different things: LMArena is people voting blind between two answers, the others are test scores. Click any row for the source.