Skip to content

status.openkey.ai is live — independent, worldwide uptime proof See it now ->

openkey

Liquid AI

3 models built

Liquid AI is an MIT CSAIL spinoff founded in 2023, built around liquid neural network architectures rather than the standard transformer stack most labs ship. The pitch is efficient, low-memory inference — models that run well on constrained hardware without giving up much capability. Their Liquid Foundation Models (LFMs) lineup on OpenKey is small, just 3 models, but it spans a clear range: a larger 24B mixture-style model (LFM2-24B-A2B) for real workloads, and two compact 1.2B instruct/thinking variants they give away for free. That split — one paid mid-size model, two free small ones — reflects a lab still proving out its architecture rather than chasing frontier benchmarks.

Lineup

For actual production work, use LFM2-24B-A2B: $0.03/1M input and $0.12/1M output on OpenKey (provider price × 1.03), with a 128K context window — cheap enough to run at volume. If you're prototyping, testing latency on constrained infra, or just want to see how a non-transformer architecture handles instruction-following, grab LFM2.5-1.2B-Instruct (free). Need step-by-step reasoning instead of straight answers? LFM2.5-1.2B-Thinking (free) is the same size class but tuned for chain-of-thought — same zero cost, smaller 32K context.

Prices per 1M tokens, flat 3% fee included.

Questions

What does Liquid AI charge for its models on OpenKey?
Two of the three models are free: LFM2.5-1.2B-Instruct and LFM2.5-1.2B-Thinking cost $0 for input and output. The paid option, LFM2-24B-A2B, runs $0.03/1M input tokens and $0.12/1M output tokens on OpenKey — that's the provider list price plus the flat 3% fee. It's one of the cheaper paid models on the platform.
Which Liquid AI model should I use for a real project?
LFM2-24B-A2B is the only paid, larger-scale option and the newest in the lineup (Feb 2026), with a 128K context window at $0.03/$0.12 per 1M input/output tokens. The two free 1.2B models are better suited to lightweight tasks or testing — they cap out at 32K context.
Does Liquid AI have free models available?
Yes — 2 of their 3 models on OpenKey are free: LFM2.5-1.2B-Instruct and LFM2.5-1.2B-Thinking, both released January 2026 with 32K context. They're small (1.2B) but cost nothing, making them a reasonable starting point before paying for LFM2-24B-A2B.