tensor.news
Integrity IndexModelsCompareLabsPapersMethodologyAbout
tensor.news

© 2026

Evidence

  • Integrity Index
  • Model Claim Ledger
  • Compare

Entities

  • Models
  • Benchmarks
  • Labs
  • Papers

Method

  • Methodology
  • About
  • Privacy

Machine

  • Feed (Atom)
  • llms.txt
  • llms-full.txt
  • Sitemap

tensor.news grades benchmarks and tags model claims by evidence status. Every fact carries a source record and provenance. Scores are framed as task performance under a disclosed harness, never as deployed capability.

← Models

DeepSeek-V3 (Mar 2025)

DeepSeek·released 2025-03-24

independently reproduced

Scored on 9 benchmarks. 33% of its scores are independently reproduced — the rest are self-reported or unverified.

9
Benchmarks scored
0
SOTA leads
77
Best score
33%
Reproduced
3 of 9

Benchmark fingerprint

Every score, ranked, colored by evidence status.

  • Lech Mazur Writing
    77
  • MATH level 5
    75.55
  • GPQA diamond
    56.82
  • Aider polyglot
    55.1
  • Fiction.LiveBench
    50
  • OTIS Mock AIME 2024-2025
    37.72
  • WeirdML
    36.08
  • SimpleBench
    12.64
  • CritPt
    0

All benchmark scores

BenchmarkScoreEvidence statusMeasured
Lech Mazur Writing77accuracy (%)unverified2025-03-24
MATH level 575.55accuracy (%)reproduced· optimized2025-03-24
GPQA diamond56.82accuracy (%)reproduced· optimized2025-03-24
Aider polyglot55.1accuracy (%)unverified· optimized2025-03-24
Fiction.LiveBench50accuracy (%)unverified2025-03-24
OTIS Mock AIME 2024-202537.72accuracy (%)reproduced· optimized2025-03-24
WeirdML36.08accuracy (%)unverified2025-03-24
SimpleBench12.64accuracy (%)unverified2025-03-24
CritPt0accuracy (%)unverified2025-03-24
9 source records · epoch_benchmarks. Every score is evidence-linked.

Follow the record

More from DeepSeek

  • DeepSeek-V4-Pro96.66
  • DeepSeek-R1 (May 2025)96.64
  • DeepSeek-V393.73
  • DeepSeek-R193.05
  • DeepSeek-V2 (MoE-236B, May 2024)89.6
  • DeepSeek-V3.287.81