Liquid LFM2.5-2.6B Card - BrowseComp+ (OpenClaw): leaderboard
BrowseComp-Plus, a fixed-corpus version of BrowseComp's hard browsing questions, answered through the OpenClaw agent harness and run by Liquid AI with vLLM for the LFM2.5-2.6B model card to compare small on-device models. Accuracy in percent, higher is better; thinking modes of the Gemma and Qwen models are not stated.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Around April 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 9B | 27.23 | #391 |
| 2 | LFM2.5-2.6B | 26.89 | #836 |
| 3 | Qwen 3.5 4B | 24.46 | #514 |
| 4 | gemma-4-E4B-it | 15.9 | #577 |
| 5 | Gemma 4 E2B (IT) | 8.31 | #779 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=liquid-lfm2-5-2-6b-card-browsecomp-plus-openclaw · How It Works · Data refreshed daily, snapshot 2026-11-05.