← All lab files

DeepSeek research file

#6 on the current provider board.

DeepSeek V4 Pro scores 1457.1 according to LMArena's text style-control snapshot published July 12, 2026.

LMArena source rank#6
LMArena rating1457.1
Reported interval1452.6–1461.6
Battles41,800

Model records from the Epoch AI catalog

DeepSeek models in the reviewed record.

Release date, domains, tasks, access, and weight availability come from the Epoch AI catalog. These fields describe models but do not rank them.

  1. DeepSeek-V4-Pro

    Domains
    Language
    Access
    Open weights (unrestricted)
    Weights
    Open
    Read profile →

  2. DeepSeekMath-V2

    Domains
    Language
    Access
    Open weights (unrestricted)
    Weights
    Open
    Read profile →
  3. Domains
    Language
    Access
    Open weights (unrestricted)
    Weights
    Open
    Read profile →

Historical momentum · legacy metric only

+416.4 from first to last sampled observation.

This change is calculated only inside the selected legacy series, from March 7, 2024 to August 29, 2025. The archive crosses methodology changes and cannot be joined to, trended into, or compared numerically with the current snapshot.

First sample1008.0deepseek-llm-67b-chat
Legacy peak1424.42025-08-29 · deepseek-r1-0528
Last sample1424.4deepseek-r1-0528
Show all 7 legacy observations
  1. deepseek-llm-67b-chat1008.0
  2. deepseek-coder-v21179.0
  3. deepseek-v2-api-06281219.2
  4. deepseek-v2.51257.5
  5. deepseek-r11361.4
  6. deepseek-v3-03241368.5
  7. deepseek-r1-05281424.4

Release record

Latest matched notable models.

Epoch AI records provide dated context only. A release record does not imply a score change.

  1. DeepSeek-V4-Flash
  2. DeepSeek-V4-Pro
  3. DeepSeekMath-V2
  4. DeepSeek-R1 (May 2025)
  5. DeepSeek-V3 (Mar 2025)

Continue researching

Profiles, guides, and reports.

Evidence boundary

Three sources. Three distinct jobs.

Current frontier: LMArena leaderboard dataset, revision 4a6c22ada142f24563a2ecc5ceb1ce94ab8be019, CC BY 4.0.

Historical context: LMArena historical Space artifacts, with one SHA-256 receipt per selected sample and no continuity to the current metric.

Catalog and releases: Epoch AI notable models. No corporate profile is added because our sources do not establish company-level facts beyond model authorship.

Read the full methodology