meta-llama/Llama-4-Maverick-17B-128E-Instruct (official Meta model card, Benchmarks section)
repo
Fields
id
sources/llama4-maverick-hf-cardKind
repo
Accessed
2026-10-19
notes
Read https://huggingface.co/meta-llama/Llama-4-Maverick-17B-128E-Instruct for the benchmark figures on models/llama-4-maverick. As recorded by the reader: Instruction-tuned model table, 0-shot, MMLU Pro = macro_avg/acc 80.5 and GPQA Diamond = accuracy 69.8 (Llama 4 Maverick column). Card also reports pre-trained-model numbers (MMLU 85.5, MMLU-Pro 62.9 5-shot) which are not used here. Card's LiveCodeBench entry is 'LiveCodeBench (10/01/2024-02/01/2025)' 43.4, not LiveCodeBench v6, so it was excluded.
Notes
Read https://huggingface.co/meta-llama/Llama-4-Maverick-17B-128E-Instruct for the benchmark figures on models/llama-4-maverick. As recorded by the reader: Instruction-tuned model table, 0-shot, MMLU Pro = macro_avg/acc 80.5 and GPQA Diamond = accuracy 69.8 (Llama 4 Maverick column). Card also reports pre-trained-model numbers (MMLU 85.5, MMLU-Pro 62.9 5-shot) which are not used here. Card's LiveCodeBench entry is 'LiveCodeBench (10/01/2024-02/01/2025)' 43.4, not LiveCodeBench v6, so it was excluded.
Cross-references
This record references no other records.