LFM2-1.2B
1.2B hybrid-architecture model built for on-device inference.
About
Liquid AI's second-generation foundation model in a 1.2B package, built on a hybrid convolution-plus-attention architecture tuned for CPUs and NPUs rather than datacenter GPUs.
Liquid claims 2x faster decode and prefill than Qwen3 on CPU and 3x better training efficiency than the first LFM generation. Note the license: Apache 2.0-based, but commercial use requires a separate agreement above $10M revenue.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
2x faster decode/prefill than Qwen3 on CPU per Liquid AI; outperforms similar-size models on knowledge, math, and instruction-following benchmarks per the company.
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (metrics tier, metrics-v1). Evidence: Growing adoption for on-device and edge deployments; 4 size variants published; Liquid AI actively ships LFM2.x follow-ups and vision variants; Apache 2.0-based license with a $10M revenue threshold for commercial use — not fully permissive. Manual review: speed claims are company-reported; context length not published. License: Apache 2.0-based with commercial revenue restriction above $10M.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 6.0/10
- Maintenance · recency, releases, issue load
- 7.0/10
- Openness · license + docs
- 6.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · Growing adoption for on-device and edge deployments; 4 size variants published
- · Liquid AI actively ships LFM2.x follow-ups and vision variants
- · Apache 2.0-based license with a $10M revenue threshold for commercial use — not fully permissive
- · manual review: speed claims are company-reported; context length not published