LFM2.5-2.6B: benchmark results
Liquid AI's LFM2.5-2.6B, a 2.69B hybrid on-device model on the LFM2 architecture with a 128K context window, trained on 34T tokens with agentic post-training for tool use. Provider: Liquid AI. Released 2026-08-04. Access: Open.
Aggregate Rating 1531 ± 34, rank #675 of 1635 rated models, from 66 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Liquid LFM2.5-2.6B Card - AA-Omniscience-Public Index | -29.5 | Omniscience Index (-100 to 100) | 100 |
| Liquid LFM2.5-2.6B Card - AA-Omniscience-Public Non-Hallucination Rate | 59.04 | Non-Hallucination Rate (%) | 100 |
| Liquid LFM2.5-2.6B Card - IFStruct | 85.49 | Pass Rate (%) | 100 |
| AA-Omniscience Hallucination Rate | 16 | Hallucination Rate (%) | 99.1 |
| LLM Stats (Multi-IF) | 80.07 | Score (%) | 91.7 |
| AA-Omniscience Index - Law | -3.6 | Omniscience Index | 78.9 |
| AA-Omniscience Index - Health | -10.7 | Omniscience Index | 77 |
| Liquid LFM2.5-2.6B Card - BrowseComp+ (OpenClaw) | 26.89 | Accuracy (%) | 75 |
| AA-Omniscience Index - Software Engineering (SWE) - R | -6 | Omniscience Index | 72.3 |
| AA-Omniscience Index - Software Engineering (SWE) - Julia | -12 | Omniscience Index | 70.1 |
| AA-Omniscience Index - Humanities & Social Sciences | -8 | Omniscience Index | 69.5 |
| AA-Omniscience Index - Business | -10.2 | Omniscience Index | 68.8 |
Interactive version: theaggregate.ai/model?slug=lfm2-5-2-6b · How It Works · Data refreshed daily, snapshot 2026-10-15.