Liquid LFM 2.5
Liquid LFM 2.5
Still not a transformer, and still the model to beat on a phone or a laptop CPU.
Updated
The read
LFM 2.5 is Liquid AI's current small family, including a 2.6-billion-parameter dense model released in August and an eight-billion sparse model that activates only one billion parameters per token. The architecture is built for devices without a GPU, and it runs fast enough on a laptop CPU to feel instant. It knows little and reasons shallowly, as models this size do. The licence is free only for companies under ten million dollars in revenue.
Where it shines
- Runs quickly on phones and laptop CPUs
- A sparse 8B variant that activates only one billion parameters
- 128,000 tokens of context at a tiny size
Worth knowing first
- The licence is free only below ten million dollars in annual revenue
- Shallow reasoning and thin world knowledge, as expected at this size
Index score
68.9
Provisional. This score is read from the product's public capability, not from the one-prompt rebuild the fully reviewed entries went through, so treat it as a placing rather than a verdict.
- Rank
- 71 of 73
- Builds
- LLMs
- Category
- Small and fast
- Output
- Small models sized for phones and edge devices
- Pricing
- Free tier
- Best for
- On-device inference with no GPU and no network
Scores are our own editorial judgement, weighted craft 55, speed 10, control 15 and value 20.
Also worth a look
Full index- 11Claude Haiku 5.5Claude Haiku 5.5ProvisionalA small model priced like a rounding error that scores where mid-tier models did last quarter.Small and fastPaid only87.9
- 22GPT-6 LunaGPT-6 LunaProvisionalThe cheap end of GPT-6, and the model free ChatGPT users now talk to.Small and fastPaid only84.9
- 30GPT-5.6 LunaGPT-5.6 LunaProvisionalStill cheap and still capable, but GPT-6 Luna costs half as much.Small and fastPaid only83.0
