LiquidAI: LFM2-24B-A2B
Liquid
Capabilities
Strengths
- Extremely cost-effective ($0.03/$0.12 per M tokens)
- 128K context window enables long-document processing
- Low inference energy via sparse MoE activation
- Open weights enable on-device deployment and full data privacy
Weaknesses
- Weak code generation (0.07 fitness)
- Poor reasoning and analysis capabilities
- Very slow inference speed (126 tok/s)
- Limited to simple summarization and extraction tasks
Pricing
Input / 1M tokens
$0.03
Output / 1M tokens
$0.12
Context window
128k
Transparency
Open weights
10.0 / 10
Open training data
0.0 / 10
Open methodology
6.5 / 10
Licence openness
5.5 / 10
Provider disclosure
5.0 / 10
FMTI company score
N/A
Open-weight model with documented architecture (MoE with 2B active parameters). Liquid provides reasonable model cards and technical documentation. Training data sources not fully disclosed. Architecture transparency is solid for an open-weight model.
Sustainability
Inference energy
7.5 / 10
Training footprint
N/A
Provider infrastructure
5.5 / 10
24B parameter MoE with only 2B active parameters significantly reduces inference energy relative to dense models of similar scale. Efficient on-device deployment design. Training footprint and provider infrastructure data not publicly available.