aya-101: benchmark results

Cohere Labs Aya 101 is an instruction-tuned multilingual encoder-decoder model covering 101 languages. It uses the mT5-XXL architecture with 24 encoder and 24 decoder layers; the publisher reports 13B parameters and 25 million fine-tuning samples under Apache 2.0. Provider: Cohere. Released 2024-02-13. Access: Open.

Aggregate Rating 1432 ± 30, rank #4695 of 10524 rated models, from 46 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
Open PL LLM - 8TAGS (multiple choice, 5-shot)76.28Accuracy (%)70.9
Open PL LLM - PolEmo2-OUT (multiple choice, 5-shot)72.67Accuracy (%)69.1
Open PL LLM - PPC (generative, 5-shot)73.2Accuracy (%)65.6
Open PL LLM - EQ-Bench35.7EQ-Bench Score63.2
Open PL LLM - 8TAGS (generative, 5-shot)73.4Accuracy (%)57.7
Open PL LLM - PolQA Reranking (multiple choice, 5-shot)72.09Accuracy (%)57.2
Open PL LLM - Belebele (generative, 5-shot)79.67Accuracy (%)54.6
Open PL LLM - Belebele (multiple choice, 5-shot)79.67Accuracy (%)53.5
Greek MMLU - Other57.05Accuracy (%)50.6
Greek MMLU - Social Sciences62.16Accuracy (%)50.6
Open PL LLM - KLEJ NER (multiple choice, 5-shot)46.06Accuracy (%)49.8
Greek MMLU56.73Accuracy (%)49.4

Interactive version: theaggregate.ai/model?slug=aya-101 · How It Works · Data refreshed daily, snapshot 2026-11-22.