LFM2.5-24B-A2B
LFM2.5-24B-A2B is a 24B-parameter text-only reasoning model with 2.45B active parameters, post-trained for agentic workloads. It supports tool calling, has a 131,072-token context length, and is released as a model checkpoint for use with Transformers, vLLM, and SGLang.
What it does
Use it to build agents that reason, call tools, and run locally on-device; for knowledge-heavy tasks, pair it with retrieval or web search.
- Availability
- Original model checkpoint in native format; available on Hugging Face
- Pricing model
- Not established
- Deployment
- Local and on-device deployment supported
- Privacy
- Built for private agents on-device
- Licence
- LFM Open License v1.0
Capability
Modality
Access
Deployment
Pricing
Target user
Evidence
10 claims- The model was converted to MLX format using mlx-lm version 0.32.0. [1]
This model [LiquidAI/LFM2.5-24B-A2B-MLX-4bit](https://huggingface.co/LiquidAI/LFM2.5-24B-A2B-MLX-4bit) was converted to MLX format from [LiquidAI/LFM2.5-24B-A2B](https://huggingface.co/LiquidAI/LFM2.5-24B-A2B) using mlx-lm version **0.32.0**.
Liquid AI. Quoted text is outside CC BY 4.0 and keeps its original rights. - The model is quantized to 4 bits per weight, with token embeddings kept at 6 bits and MoE routing gates kept unquantized in float32. [1]
- Loading the model requires mlx-lm 0.32.0 or later. [1]
Requires mlx-lm 0.32.0 or later.
Liquid AI. Quoted text is outside CC BY 4.0 and keeps its original rights. - Earlier mlx-lm versions load the model but route the experts incorrectly. [1]
Earlier versions load the model but route the experts incorrectly.
Liquid AI. Quoted text is outside CC BY 4.0 and keeps its original rights. - The model has 24B total parameters and 2.45B active parameters. [2]
- **Best-in-class agent**: Competitive with Qwen3.6-35B-A3B, a model with ~50% more total and active parameters, on agentic and tool-use tasks, scoring 48.09 on LiveClaw versus 38.19.
Hugging Face. Quoted text is outside CC BY 4.0 and keeps its original rights. - The model has a context length of 131,072 tokens. [2] [3]
- **Context length**: 131,072 tokens
Hugging Face. Quoted text is outside CC BY 4.0 and keeps its original rights. - LFM2.5-24B-A2B supports function calling for agentic workflows. [2]
LFM2.5-24B-A2B supports tool calling for agentic workflows.
Hugging Face. Quoted text is outside CC BY 4.0 and keeps its original rights. - The original model checkpoint is available in native format on Hugging Face. [2]
| **[LFM2.5-24B-A2B](https://huggingface.co/LiquidAI/LFM2.5-24B-A2B)** | Original model checkpoint in native format.
Hugging Face. Quoted text is outside CC BY 4.0 and keeps its original rights. - The model can be run locally with inference frameworks including Transformers, vLLM, SGLang, and llama.cpp. [2]
- The source recommends pairing the model with retrieval or web search for knowledge-heavy tasks. [2] [3]
It is not recommended for knowledge-heavy tasks on its own: pair it with retrieval or web search, which plays to its tool-calling strength.
Hugging Face. Quoted text is outside CC BY 4.0 and keeps its original rights.
Sources
3
1
Liquid AIOwner-controlled
https://huggingface.co/LiquidAI/LFM2.5-24B-A2B-MLX-4bit
Retrieved 2026-10-27 01:17 UTC
Verified
2
Hugging FaceOwner-controlled
https://huggingface.co/LiquidAI/LFM2.5-24B-A2B
Retrieved 2026-10-27 01:17 UTC
Verified
3
Verified
Versions
11| Version | Effective | Change | Changed fields | State |
|---|---|---|---|---|
| 11 | 2026-10-27 01:17 UTC | material change | capability, modality, access, deployment, availability, deployment | Current |
| 10 | 2026-10-26 00:38 UTC | material change | capability, modality, access, deployment, availability, deployment | Current |
| 9 | 2026-10-25 00:36 UTC | material change | capability, modality, access, deployment, availability, deployment | Current |
| 8 | 2026-10-24 01:06 UTC | material change | capability, modality, access, deployment, availability, deployment | Current |
| 7 | 2026-10-23 00:46 UTC | material change | capability, modality, access, deployment, availability, deployment | Current |
| 6 | 2026-10-22 00:04 UTC | material change | capability, access, deployment, availability, deployment | Current |
| 5 | 2026-10-22 00:04 UTC | material change | capability, modality, access, availability, deployment | Current |
| 4 | 2026-10-20 15:57 UTC | material change | capability, access, deployment, availability, deployment | Current |
| 3 | 2026-10-19 16:57 UTC | material change | capability, access, deployment, availability, deployment | Current |
| 2 | 2026-10-14 21:24 UTC | material change | capability, modality, access, availability, deployment, privacy | Current |
| 1 | 2026-10-13 22:16 UTC | publication | None | Current |
In briefs
1- An open model for tool-using agents, plus a checkpoint to fine-tune Daily brief 2026-10-13
Original catalog content is licensed under CC BY 4.0, attributed to wowsuchai. Source quotations and linked content keep their original rights.
