LFM3 on BenchGecko · 1.7B parameters, 262,144 context
Liquid AI released the LFM3 family on 2026-10-27 with open weights. BenchGecko lists four models, led by LFM3-1.7B with a 262,144 token context.
What launched
Liquid AI released the LFM3 family on 2026-10-27 with open weights on Hugging Face under its LFM 1.0 license. BenchGecko added four models to the Liquid AI provider page today.
- LFM3-1.7B · text model, about 1.7B parameters, 262,144 token context
- LFM3-VL-2B · image and video input, about 1.9B parameters, 128,000 token context
- LFM3-Audio-2B · audio in and out, about 2B parameters, 18 languages
- LFM3-Omni-2B · vision and audio input, about 2.2B parameters, 128,000 token context
A base checkpoint, LFM3-1.7B-Base, and two decision models, d1-2B and d1-Omni-2B, shipped too and are not listed separately.
Pricing on BenchGecko
No provider lists a hosted price for LFM3 yet, and the family is not on OpenRouter. BenchGecko leaves the USD price fields empty until a provider publishes one.
LFM3-1.7B against LFM2.5-2.6B
LFM2.5-2.6B is already on BenchGecko at $0.35 input and $0.50 output per million tokens with a 128,000 token context. The new 1.7B model is about 35% smaller and doubles the context to 262,144 tokens.
Liquid AI reports these scores in the same launch table. BenchGecko records them as vendor claims, not independent results.
- GPQA Diamond · 63.37 for LFM3-1.7B · 57.78 for LFM2.5-2.6B
- IFEval · 96.66 · 95.39
- BFCL V4 · 71.39 · 55.89
- SWE-bench Verified with OpenCode · 43.40 · 18.00
- RULER at 128K · 65.56 · 1.15
Other vendor-reported LFM3-1.7B scores: IFBench 73.03, LiveCodeBench v6 73.46, AA-LCR 52.83, MCP-Atlas 32.57, WebArena-Verified 38.42, SciCode 28.47.
Sources
Source: BenchGecko
Written automatically from BenchGecko's tracked data for this period; every number comes from the changelog, model, mindshare and pricing datasets. Checked against the source data before publishing.