LFM2.5-1.2B Instruct: benchmark results
Liquid AI's LFM2.5-1.2B-Instruct, fine-tuned from LFM2.5-1.2B-Base with pretraining scaled to 28 trillion tokens and large-scale RL, using a 16-layer hybrid convolution/attention architecture and 32,768-token context. Provider: Liquid AI. Released 2026-01-05. Access: Open.
Aggregate Rating 1308 ± 17, rank #1462 of 1656 rated models, from 405 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| MERA Code - CodeCorrectness | 84.5 | Accuracy (%) | 89.2 |
| Kurakurai French LLM - IFEval | 72.64 | Accuracy (%) | 82.9 |
| Kurakurai French LLM - GPQA Diamond | 29.9 | Accuracy (%) | 80.5 |
| Kurakurai French LLM - MultiIF | 53.66 | Accuracy (%) | 80.5 |
| Kurakurai French LLM - MATH-500 | 63.88 | Accuracy (%) | 75.6 |
| Kurakurai French LLM - AIME 2025 | 5.33 | Accuracy (%) | 72 |
| Kurakurai French LLM - AIME 2024 | 4.33 | Accuracy (%) | 70.7 |
| Kurakurai French LLM - MBPP | 48.65 | pass@1 (%) | 70.7 |
| Kurakurai French LLM - HumanEval+ | 44.19 | pass@1 (%) | 68.3 |
| Kurakurai French LLM - MGSM | 65.68 | Accuracy (%) | 68.3 |
| Kurakurai French LLM - MMLU-Pro | 37.4 | Accuracy (%) | 68.3 |
| Kurakurai French LLM - Global MMLU | 54.9 | Accuracy (%) | 63.4 |
Interactive version: theaggregate.ai/model?slug=lfm2-5-1-2b-instruct · How It Works · Data refreshed daily, snapshot 2026-10-20.