Benchmark

Meta Llama — reported benchmarks

Meta · source page ↗ · last checked Oct 20, 2026, 12:03 AM

Reported benchmarks · Llama 4 Maverick

captured Oct 20, 2026, 12:03 AM
BenchmarkScore
MMLU85.5% · instruction-tuned
MMLU Pro80.5% · instruction-tuned
GPQA Diamond69.8% · instruction-tuned
LiveCodeBench43.4% pass@1 · instruction-tuned
AIME 202539.0% · instruction-tuned
MMMU73.4% · image reasoning; instruction-tuned
MathVista73.7% · image reasoning; instruction-tuned
ChartQA90% · image understanding; instruction-tuned
DocVQA94.4% · image understanding test split; instruction-tuned
LMArena Elo1417 Elo · experimental chat variant; vendor-claimed
Terminal-Bench 2.17.9% · agentic task success
Terminal-Bench Hard6.8% · agentic task success

Vendor-reported via automated web search — not independently verified. See the cited matrix on /models.

Change history