BenchLeader
Menu
LeaderboardsModelsProvidersBenchmarksCompareReleasesBlogMethodology
DeepSeekOpen weights: deepseek-ai/DeepSeek-V3

DeepSeek V3

Best configuration ranks #433 of 610 on the BenchLeader Index at 43.8 ±3.5. Last measured 2 Sept 2026. Released 26 Dec 2024.

Blended price
$0.478/M
$0.270 in · $1.10 out
Output speed
–
First answer
–
Context
66k
How it scores by categoryDashed line = average model (50). One step of 15 = one standard deviation.
  1. Overall index44
  2. Reasoning39
  3. Coding50
  4. Agents & tools40
  5. Maths43
  6. Knowledge44
  7. Instruction following38
  8. Human preference53
  9. Long context40
  10. Composite40

Benchmark results

One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.

Reasoning

BenchmarkdefaultSourceTrend
GPQA Diamond56.5%#185Epoch AI Benchmarking Hub
SimpleBench18.9%#80SimpleBench
LMArena Hard Prompts1350#193LMArena
GPQA Diamond (AA)not in index55.7%#383Artificial Analysis
Humanity's Last Exam (AA)not in index2.9%#516Artificial Analysis
GPQA Diamond (Vals)not in index54.5%#112Vals AI
Kagi LLM Benchmark52.3%#73Kagi LLM Benchmark

Coding

BenchmarkdefaultSourceTrend
SciCode35.4%#133SciCode
LMArena Coding1387#183LMArena
SciCode (AA)not in index35.8%#137Artificial Analysis
Aider Polyglot55.1%#18Aider polyglot leaderboard

Agents & tools

BenchmarkdefaultSourceTrend
Terminal-Bench Hard6.8%#240Artificial Analysis
τ²-Bench Telecom (AA)not in index22.8%#295Artificial Analysis

Maths

BenchmarkdefaultSourceTrend
OTIS Mock AIME15.8%#207Epoch AI Benchmarking Hub
MATH Level 564.8%#42Epoch AI Benchmarking Hub
AIME (Vals)27.5%#72Vals AI
MGSM92.2%#21Vals AI

Knowledge

BenchmarkdefaultSourceTrend
AA-Omniscience-41.6#262Artificial Analysis
MMLU-Pro73.8%#110Vals AI
LegalBench80.8%#76Vals AI
CorpFin52.5%#97Vals AI
TaxEval67.9%#101Vals AI
MedQA80.9%#68Vals AI

Instruction following

BenchmarkdefaultSourceTrend
IFBench34.8%#298Artificial Analysis

Human preference

BenchmarkdefaultSourceTrend
LMArena Text1358#170LMArena

Long context

BenchmarkdefaultSourceTrend
AA-LCR29.3%#325Artificial Analysis

Composite

BenchmarkdefaultSourceTrend
Epoch Capabilities Indexnot in index135.9#114Epoch AI Benchmarking Hub
AA Intelligence Index8.5#359Artificial Analysis

Price history

Listed price per 1M tokens over time, as recorded by OpenRouter for the provider with the longest history.

$0.00$0.248$0.495$0.743$0.990Jan 26Feb 26Mar 26Apr 26May 26Jun 26Jul 26Aug 26
input outputnow $0.240 in · $0.900 out

What a task costs

Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache. Reasoning tokens are not modelled.

WorkloadTokens in / outCostWith cachingTime
Chat reply400 / 300$0.0004––
Summarise a 30-page report12,000 / 600$0.0039––
Code edit6,000 / 1,500$0.0033––
Agentic coding session60,000 / 4,000$0.021––
Structured extraction2,000 / 200$0.0008––

See also

Best under $2/MBest under $0.50/MBest open weightsBest for codingAll DeepSeek models

Data as of 9 Sept 2026. Compare with another model.