Tihu · Pelican Bench

Claude Sonnet 5.5

Anthropic · Model release:

AI reference score 6.89 ± 0.98 · high (best) Last 4 weekly runs · 12 images
Tier comparison
TierSamplesAI reference scoreValid SVG rateMean timeSamples
high
Claude Sonnet 5.5 · high · 2026-10-25 · #3 · 6.89
 6.89 ± 0.98Last 4 weekly runs · 12 images 100%35 s3
low
Claude Sonnet 5.5 · low · 2026-10-25 · #1 · 6.38
 5.84 ± 0.66Last 4 weekly runs · 12 images 100%18 s3
medium default
Claude Sonnet 5.5 · medium · 2026-10-25 · #2 · 6.46
 5.79 ± 0.77Last 4 weekly runs · 12 images 100%21 s3
Claude Sonnet 5.5 · high · 2026-10-25 · #3 · 6.89
high 6.89 ± 0.98Last 4 weekly runs · 12 images
Valid SVG rate 100% · Mean time 35 s
Claude Sonnet 5.5 · low · 2026-10-25 · #1 · 6.38
low 5.84 ± 0.66Last 4 weekly runs · 12 images
Valid SVG rate 100% · Mean time 18 s
Claude Sonnet 5.5 · medium · 2026-10-25 · #2 · 6.46
medium default5.79 ± 0.77Last 4 weekly runs · 12 images
Valid SVG rate 100% · Mean time 21 s

Latest samples

high

AI reference score 6.89 ± 0.98 Last 4 weekly runs · 12 images Monitoring
Top tiers of one modelView tier page

SunLatest

Claude Sonnet 5.5 · high · 2026-10-25 · #3 · 6.89Claude Sonnet 5.5 · high · 2026-10-25 · #1 · 6.81Claude Sonnet 5.5 · high · 2026-10-25 · #2 · 6.19

low

AI reference score 5.84 ± 0.66 Last 4 weekly runs · 12 images Monitoring
Top tiers of one modelView tier page

SunLatest

Claude Sonnet 5.5 · low · 2026-10-25 · #1 · 6.38Claude Sonnet 5.5 · low · 2026-10-25 · #3 · 5.54Claude Sonnet 5.5 · low · 2026-10-25 · #2 · 5.20

medium

AI reference score 5.79 ± 0.77 Last 4 weekly runs · 12 images Monitoring
Top tiers of one modelView tier page

SunLatest

Claude Sonnet 5.5 · medium · 2026-10-25 · #2 · 6.46Claude Sonnet 5.5 · medium · 2026-10-25 · #3 · 5.50Claude Sonnet 5.5 · medium · 2026-10-25 · #1 · 4.85

Score trend

high Normal

2026-10-08 · 7.692026-10-09 · 7.692026-10-10 · 7.692026-10-11 · 6.542026-10-12 · 6.542026-10-13 · 6.542026-10-14 · 6.542026-10-15 · 6.542026-10-16 · 6.542026-10-17 · 6.542026-10-18 · 6.702026-10-19 · 6.702026-10-20 · 6.702026-10-21 · 6.702026-10-23 · 6.702026-10-24 · 6.702026-10-25 · 6.632026-10-26 · 6.63

18 records in the last 30 days

2026-10-26 · Normal2026-10-25 · Normal2026-10-24 · Normal2026-10-23 · Normal2026-10-21 · Normal2026-10-20 · Normal2026-10-19 · Normal2026-10-18 · Collecting earlier results2026-10-17 · Normal2026-10-16 · Normal2026-10-15 · Normal2026-10-14 · Normal2026-10-13 · Normal2026-10-12 · Normal2026-10-11 · Collecting earlier results2026-10-10 · Normal2026-10-09 · Normal2026-10-08 · Collecting earlier results

low Normal

2026-10-08 · 6.552026-10-09 · 6.552026-10-10 · 6.552026-10-11 · 5.682026-10-12 · 5.682026-10-13 · 5.682026-10-14 · 5.682026-10-15 · 5.682026-10-16 · 5.682026-10-17 · 5.682026-10-18 · 5.412026-10-19 · 5.412026-10-20 · 5.412026-10-21 · 5.412026-10-23 · 5.412026-10-24 · 5.412026-10-25 · 5.702026-10-26 · 5.70

18 records in the last 30 days

2026-10-26 · Normal2026-10-25 · Normal2026-10-24 · Normal2026-10-23 · Normal2026-10-21 · Normal2026-10-20 · Normal2026-10-19 · Normal2026-10-18 · Collecting earlier results2026-10-17 · Normal2026-10-16 · Normal2026-10-15 · Normal2026-10-14 · Normal2026-10-13 · Normal2026-10-12 · Normal2026-10-11 · Collecting earlier results2026-10-10 · Normal2026-10-09 · Normal2026-10-08 · Collecting earlier results

medium Normal

2026-10-08 · 6.302026-10-09 · 6.302026-10-10 · 6.302026-10-11 · 5.892026-10-12 · 5.892026-10-13 · 5.892026-10-14 · 5.892026-10-15 · 5.892026-10-16 · 5.892026-10-17 · 5.892026-10-18 · 5.362026-10-19 · 5.362026-10-20 · 5.362026-10-21 · 5.362026-10-23 · 5.362026-10-24 · 5.362026-10-25 · 5.602026-10-26 · 5.60

18 records in the last 30 days

2026-10-26 · Normal2026-10-25 · Normal2026-10-24 · Normal2026-10-23 · Normal2026-10-21 · Normal2026-10-20 · Normal2026-10-19 · Normal2026-10-18 · Collecting earlier results2026-10-17 · Normal2026-10-16 · Normal2026-10-15 · Normal2026-10-14 · Normal2026-10-13 · Normal2026-10-12 · Normal2026-10-11 · Collecting earlier results2026-10-10 · Normal2026-10-09 · Normal2026-10-08 · Collecting earlier results

Earlier samples

high

low

medium

Versions by score

  1. Claude Sonnet 5.5
    Claude Sonnet 5.5 · high
    high 6.89Last 4 weekly runs · 12 images

Scores by criterion

Scores by tier
Scoring criteria high low medium
Pelican likeness7.927.087.58
Bicycle correctness6.504.924.92
Riding pose5.255.083.83
Composition7.256.426.75
Obvious errors (higher is better)5.924.834.67

Release notes

Original (Chinese)官方文档标注 Released September 28, 2026；当天在 Claude.ai、Claude API（claude-sonnet-5-5）及 AWS/Google Cloud/Microsoft Foundry 可用。

· View release source ↗