flan-t5-base: benchmark results

Provider: Google. Released 2022-10-21. Access: API.

Unified ELO 1335 ± 1, rank #1294 of 1392 rated models, from 13 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
ConvRe - Text2Re Hard50.2Accuracy (%)88.9
ConvRe - Re2Text Easy84.6Accuracy (%)55.6
InstructEval - Problem Solving31.6Average (%, MMLU/BBH/DROP/CRASS/HumanEval)53.1
InstructEval - Alignment (HHH)59Average (%, harmless/helpful/honest)44
ConvRe50.8Average Score (%)22.2
ConvRe - Re2Text Hard17Accuracy (%)22.2
Open LLM Leaderboard - BBH11.34Score20.9
Open LLM Leaderboard - MuSR3.22Score15.4
Open LLM Leaderboard - MMLU-Pro3.97Score11.4
ConvRe - Text2Re Easy51.2Accuracy (%)11.1
Open LLM Leaderboard - IFEval18.91Score10.4
Open LLM Leaderboard - MATH Level 51.06Score7.6

Interactive version: theaggregate.ai/model?slug=flan-t5-base · How It Works · Data refreshed daily, snapshot 2026-09-05.