Tihu · Pelican Bench

DeepSeek V4.1 Flash

DeepSeek · Model release:

AI reference score 7.54 ± 0.78 Last 2 weekly runs · 6 images 2 weekly tests leftUpdated every Sunday · 2026-10-25Top tiers of one model

Latest samples

SunLatest

DeepSeek V4.1 Flash · default · 2026-10-25 · #1 · 8.53DeepSeek V4.1 Flash · default · 2026-10-25 · #3 · 8.22DeepSeek V4.1 Flash · default · 2026-10-25 · #2 · 7.87

Score trend

default · This model version has no selectable reasoning tiers Normal

2026-10-18 · 6.882026-10-23 · 6.882026-10-24 · 6.882026-10-25 · 8.202026-10-26 · 8.20

5 records in the last 30 days

2026-10-26 · Normal2026-10-25 · Collecting earlier results2026-10-24 · Normal2026-10-23 · Normal2026-10-18 · Collecting earlier results

Earlier samples

default · This model version has no selectable reasoning tiers

Sun

DeepSeek V4.1 Flash · default · 2026-10-18 · #3 · 7.13DeepSeek V4.1 Flash · default · 2026-10-18 · #2 · 6.94DeepSeek V4.1 Flash · default · 2026-10-18 · #1 · 6.57

Versions by score

  1. DeepSeek V4.1 Flash
    DeepSeek V4.1 Flash · default
    default · This model version has no selectable reasoning tiers 7.54Last 2 weekly runs · 6 images

Scores by criterion

Pelican likeness
8.61
Bicycle correctness
8.39
Riding pose
8.06
Composition
8.11
Obvious errors (higher is better)
7.50

Release notes

Original (Chinese)当天在 DeepSeek API 可用。2026-10-17 更新为 DeepSeek-V4.1-Flash-1017：https://api-docs.deepseek.com/updates/ 。

· View release source ↗