Liquid LFM2.5: an on-device model, not a data-centre one
Liquid
Services found: 1
| # | AI model | Capabilities | API $/1M (in / out) | Context window | Price / mo | Output $/1M | Overview | Visit site |
|---|---|---|---|---|---|---|---|---|
| 1 |  LFM2.5-2.6B Liquid |    | No public price list | 65.5K tokens | API only | View |
Browse by need
Liquid AI contributes LFM2.5-2.6B, a small model from a lab working on architectures other than the standard transformer, aimed at running efficiently on constrained hardware.
The use case is edge: phones, embedded systems, offline assistants. At 2.6B parameters it is not competing with hosted flagships; it competes with not being able to run a model at all.
Edge-firstBuilt for constrained hardware, not datacentres.
Alternative architectureLiquid work targets efficiency beyond standard transformers.
Very small2.6B parameters.
Models from this vendor
1 Liquid model in this rating. Prices are input and output per 1M tokens from the vendor official price list; for dirhams multiply by 3.6725. Tap a card to open the model page.
How to choose
- Benchmark against a comparable model from a larger vendor before committing production traffic — line-ups this size get less independent testing.
- No model here has a public per-token rate; cost depends on where you host it.
- Check the context window against your longest input — the largest here is LFM2.5-2.6B at 65.5K tokens.
- Confirm support terms and deprecation policy, which smaller vendors document less thoroughly.
Common mistakes
Assuming parity with the majorsOn the hardest tasks the frontier labs remain ahead; compare on your own cases.
Skipping the exit planA smaller vendor can change terms or availability faster. Identify a fallback model before you depend on this one.
Who this suits
Cost-driven teamsLFM2.5-2.6B has no public per-token rate, so cost depends on where it is hosted.
EvaluatorsA useful comparison point when benchmarking the single model against larger line-ups.
Regional usersRelevant where this vendor platform or language coverage matters more than raw capability.
Related filters
Text & chat228 models for text generation and chat
Reasoning169 models that think before answering
Tool calling166 models that call your functions
MultilingualAll 239 models work beyond English
Alibaba (Qwen)34 models in this rating
OpenAI30 models in this rating
Google DeepMind17 models in this rating
Frequently asked questions
What can a 2.6B model do?
Classification, extraction, simple assistance — locally and without a network call.
Is it open weight?
Liquid has published models for local use.
Why not use a hosted flash model?
Latency, privacy and offline operation.
How many Liquid models are in this rating?
1, covering text & chat, reasoning, tool calling.
Is Liquid a good default choice?
For most teams no — start with a mainstream vendor and use this one where its pricing or specialisation gives a concrete advantage.
Frequently searched
By capabilityReasoningCodingVisionLong context 1M+Free tierOpen weightsLow latencyRealtime voiceImage generation
By vendorAnthropic (13)OpenAI (30)Google DeepMind (17)xAI (8)Meta (6)DeepSeek (9)Mistral AI (11)Alibaba (Qwen) (34)Z.ai (12)Moonshot AI (5)
Rate this page
Help us improve