InfiniteBench ↗2.8 points
Llama-3.2-3B-Instruct · InfiniteBench · InfiniteBench aggregate · HELMET paper · original-author harness ↗
Meta / MODEL EVIDENCE
Llama 3.2-3B-Instruct
2 benchmarks · 7 observations · 2 sources
Scores at a glance
2 benchmarks · large dot: Llama 3.2-3B-Instruct; small dots: closest models where reported
LongMemEval ↗0.8%
Llama-3.2-3B-Instruct · LongMemEval · LongMemEval S · full history direct · QA accuracy · additional-model comparison ↗
5 more reported results
1% · Llama-3.2-3B-Instruct · LongMemEval · LongMemEval S · full history CoN · QA accuracy · additional-model comparison · LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory ↗
46.6% · Llama-3.2-3B-Instruct · LongMemEval · LongMemEval S · key=value · QA accuracy · additional-model comparison · LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory ↗
50.8% · Llama-3.2-3B-Instruct · LongMemEval · LongMemEval S · fact-expanded keys · QA accuracy · additional-model comparison · LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory ↗
46.6% · Llama-3.2-3B-Instruct · LongMemEval · LongMemEval M · key=value · QA accuracy · additional-model comparison · LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory ↗
47% · Llama-3.2-3B-Instruct · LongMemEval · LongMemEval M · fact-expanded keys · QA accuracy · additional-model comparison · LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory ↗
Closest models by score
Similar positions on shared benchmark leaderboards. Smaller gaps are closer.
2 of 2 shared benchmark comparisons
How similarity is measured
Mean absolute gap in percentile positions on shared benchmark settings; lower is closer. At least 60% overlap is required, with a minimum of three when available. Independent evaluations are used when there are at least three; otherwise published report comparisons are used. Repeated model observations within one setting are averaged before ranking. Writing-style traits and composite indices are excluded. This describes similarity, not an overall quality ranking.
Sources
Every plotted observation links to its evaluation. Reported configurations, benchmark versions and task subsets can differ; similarity describes shared scores, not training lift.
- HELMET: How to Evaluate Long-Context Language Models Effectively and Thoroughly — published result table ↗
All 8 authors
Howard Yen, Tianyu Gao, Minmin Hou, Ke Ding, Daniel Fleischer, Peter Izsak, Moshe Wasserblat, Danqi Chen
- LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory — published result table ↗
Names used in source tables
Llama-3.2-3B-Instruct