LFM2 24B A2B: benchmark results
Liquid AI's LFM2-24B-A2B, the largest model in the LFM2 hybrid family, a 24B MoE model with 2B active parameters per token, for agentic tool use without reasoning traces. Provider: Liquid AI. Released 2026-02-24. Access: Open.
Aggregate Rating 1353 ± 19, rank #1379 of 1656 rated models, from 170 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| QIMMA - ArabLegalQA | 71.82 | BERTScore F1 (%) | 86.3 |
| QIMMA - ArabCulture | 68.83 | Accuracy (%, log-likelihood multiple choice) | 82.9 |
| QIMMA - MedArabiQ QA | 53.6 | BERTScore F1 (%) | 81.7 |
| MERA Code - CodeCorrectness | 81.26 | Accuracy (%) | 78.5 |
| QIMMA - 3LM HumanEval+ | 70.12 | pass@1 (%) | 74.4 |
| QIMMA - FannOrFlop | 52 | F1 (%) | 72 |
| QIMMA - Overall | 58.68 | Mean score across 14 benchmarks (%) | 72 |
| QIMMA - AraTrust | 83.91 | Accuracy (%, log-likelihood multiple choice) | 67.7 |
| MERA - BPS | 97 | Accuracy (%) | 65.2 |
| QIMMA - 3LM MBPP+ | 57.14 | pass@1 (%) | 63.4 |
| AA-Omniscience Hallucination Rate | 69.03 | Hallucination Rate (%) | 61.8 |
| QIMMA - 3LM STEM | 78.99 | Accuracy (%, log-likelihood multiple choice) | 61 |
Interactive version: theaggregate.ai/model?slug=lfm2-24b-a2b · How It Works · Data refreshed daily, snapshot 2026-10-20.