AI benchmark results used by Boîte À Out's

Arena / LMArena
Source: https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset
Licence: Creative Commons Attribution 4.0 International
Licence text and warranty disclaimer: https://creativecommons.org/licenses/by/4.0/legalcode.en
Dataset creators: Arena / LMArena and contributors.
We select text_style_control, latest split, overall category, retain the published
score and uncertainty, and associate explicitly reviewed variants with our price
catalogue. Selection, identifiers, layout and French explanations are adapted by
Boîte À Out's. Scores are not recomputed. The dataset revision, export hash and
retrieval/publication dates are retained in the distributed data.

Berkeley Function Calling Leaderboard
Source: https://github.com/HuanzhiMao/BFCL-Result
Authors: Huanzhi Mao and contributors.
Licence: Apache License 2.0, complete copy at /licenses/ai-benchmarks-bfcl.txt
We select Overall Acc from dated score/data_overall.csv archives and native FC
variants, preserving scores and archive dates. The transformed JSON, model
correspondences and display are created by Boîte À Out's.

EvalPlus
Source: https://github.com/evalplus/evalplus.github.io
Authors: EvalPlus and contributors.
Licence: Apache License 2.0, complete copy at /licenses/ai-benchmarks-evalplus.txt
We select HumanEval+ and MBPP+ pass@1 from results.json as independent metrics.
The transformed JSON, exact version correspondences and display are created by
Boîte À Out's. Unknown measurement dates remain unknown.

Only aggregate scores and their provenance are redistributed. Benchmark prompts,
model responses and third-party model weights are not imported. Names identify
the models and sources; no affiliation, sponsorship or endorsement is asserted.
Source licences and their warranty disclaimers remain applicable. Our scenario
costs are separate simulations and are not benchmark execution costs.