Skip to content

MiniMax M3.1 Flash

MiniMaxOpenCodeReasoning: high

MiniMax M3.1 Flash gets fewer than 1 vocabulary question in 10 right. It does best in Kabyle. Its weakest spot is writing Tarifit words in Tifinagh. It still makes mistakes swapping Tifinagh and Latin letters (56.5 out of 100).

Rank 9 of 9

Compare with others

TifinaghBench score

1.4out of 100

95% interval 1.2–1.7

1.4 with strict spelling

0–100, with 95% interval

Script score

56.5out of 100

95% interval 54.8–58.3

48.5 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
1.62 correct, 28 partial, 970 wrong of 1000
English → Tifinagh
0.22 correct, 0 partial, 998 wrong of 1000
Tifinagh ↔ Latin
71.0710 correct, 0 partial, 290 wrong of 1000

Tarifit

Tifinagh → English
2.618 correct, 15 partial, 960 wrong of 993
English → Tifinagh
0.11 correct, 0 partial, 992 wrong of 993
Tifinagh ↔ Latin
41.7417 correct, 0 partial, 583 wrong of 1000

Kabyle

Tifinagh → English
3.217 correct, 29 partial, 951 wrong of 997
English → Tifinagh
0.99 correct, 0 partial, 988 wrong of 997
Tifinagh ↔ Latin
56.9569 correct, 0 partial, 431 wrong of 1000

By dialect

Meaning tasks only

  • Standard Moroccan Amazigh0.9
  • Tarifit1.3
  • Kabyle2.0

By task

 

  • Tifinagh → English2.4
  • English → Tifinagh0.4
  • Tifinagh ↔ Latin56.5

Tifinagh → English grades

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

  • Correct
  • Partial
  • Wrong
  • Failed
  • Standard Moroccan Amazigh2 correct, 28 partial, 970 wrong of 1000

  • Tarifit18 correct, 15 partial, 960 wrong of 993

  • Kabyle17 correct, 29 partial, 951 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 Amazigh89.6%9.6%
Tarifit97.6%28.9%
Kabyle85.6%9.8%

The run

Setup
Official setup
Harness
OpenCode 2.0.32@opencode-ai/client 0.0.0-beta-19271
Judge
Claude Haiku 5.5 (Claude Code)
Sample
1,000 words per dialect, seed 1
Batch size
100
Sessions
96 answer, 30 judge
Tokens
223,193 in · 144,977 out · 2,268,884 reasoning
Model cost
$1.48
Judge cost
$0.20
Dates
24 Oct 2026, 22:48 UTC → 25 Oct 2026, 00:14 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