Skip to content

Claude Sonnet 5.5

AnthropicClaude CodeReasoning: high

Claude Sonnet 5.5 gets about 1 vocabulary question in 10 right. It does best in Kabyle. Its weakest spot is writing Standard Moroccan Amazigh words in Tifinagh. It swaps Tifinagh and Latin letters well (89.4 out of 100).

Rank 6 of 9Tied with GLM 5.3 Flash

Compare with others

TifinaghBench score

12.9out of 100

95% interval 12.1–13.6

12.3 with strict spelling

0–100, with 95% interval

Script score

89.4out of 100

95% interval 88.3–90.5

73.6 with strict spelling

Transliteration score, 0–100

Scores by dialect and task

Share of the score earned in each cell, 0–100. The transliteration column measures the script, not the vocabulary.

Standard Moroccan Amazigh

Tifinagh → English
9.467 correct, 53 partial, 880 wrong of 1000
English → Tifinagh
0.99 correct, 0 partial, 991 wrong of 1000
Tifinagh ↔ Latin
97.4974 correct, 0 partial, 26 wrong of 1000

Tarifit

Tifinagh → English
14.1104 correct, 72 partial, 817 wrong of 993
English → Tifinagh
2.222 correct, 0 partial, 971 wrong of 993
Tifinagh ↔ Latin
78.0780 correct, 0 partial, 220 wrong of 1000

Kabyle

Tifinagh → English
38.6326 correct, 117 partial, 554 wrong of 997
English → Tifinagh
12.0120 correct, 0 partial, 877 wrong of 997
Tifinagh ↔ Latin
92.8928 correct, 0 partial, 72 wrong of 1000

By dialect

Meaning tasks only

  • Standard Moroccan Amazigh5.1
  • Tarifit8.2
  • Kabyle25.3

By task

 

  • Tifinagh → English20.7
  • English → Tifinagh5.1
  • Tifinagh ↔ Latin89.4

Tifinagh → English grades

Free-text meanings, graded by the judge or matched exactly.

  • Correct
  • Partial
  • Wrong
  • Failed
  • Standard Moroccan Amazigh67 correct, 53 partial, 880 wrong of 1000

  • Tarifit104 correct, 72 partial, 817 wrong of 993

  • Kabyle326 correct, 117 partial, 554 wrong of 997

Character error rate

For exact-match answers: the smallest edit distance to an accepted answer, divided by its length. Lower is better; it shows near misses that a 0 score hides.

English → TifinaghTifinagh ↔ Latin
Standard Moroccan Amazigh85.9%1.2%
Tarifit91.4%18.1%
Kabyle67.9%2.5%

The run

Setup
Official setup
Harness
Claude Code 2.1.302@anthropic-ai/claude-agent-sdk 0.3.303
Judge
Claude Haiku 5.5 (Claude Code)
Sample
1,000 words per dialect, seed 1
Batch size
100
Sessions
97 answer, 30 judge
Tokens
736,459 in · 477,488 out · 238,873 reasoning
Model cost
$5.98 (API-equivalent; run on a subscription)
Judge cost
$0.22
Dates
20 Oct 2026, 15:40 UTC → 20 Oct 2026, 15:56 UTC
CLI version
0.1.0
Dataset checksums

SHA-256 of each dictionary file, so the run can be checked once the datasets are released.

  • Standard Moroccan Amazigh633505d9ccd2244f9a53b64f56bd153c428cc3dbd7229066845404bf3d7c5060
  • Tarifitd144a5619aa31cc06ffcc4eb26c361c65ca2c71451dfdfbea7fa134206481c4f
  • Kabyleed3a9882ac0f4ef10630917a302fd7d9ff6ac706654bb317b94c715f6b4cf9ea

Run folder on GitHub