LFM2.5-24B-A2B-Base

Developers and researchers can fine-tune it for language- or domain-specific assistants, proprietary-data training, or experiments with post-training approaches.

Availability

Checkpoint available on Hugging Face

Pricing model

Not established

Deployment

Designed for on-device deployment

Privacy

Not established

Licence

license: other

• LFM2.5-24B-A2B-Base is a pre-trained text-only checkpoint. ^[1]

  LFM2.5-24B-A2B-Base is the pre-trained text-only checkpoint, used to create all the LFM2.5-24B-A2B variants.

• The model has 24.0B total parameters and 2.45B active parameters. ^[1]

  LFM2.5-24B-A2B is the largest model in the family: a 24B-parameter mixture of experts with only 2B active parameters.

• The model has a context length of 131,072 tokens. ^[1]

  - **Context length**: 131,072 tokens

• The checkpoint is available at LiquidAI/LFM2.5-24B-A2B-Base on Hugging Face. ^{[1] [2]}

  LFM2.5-24B-A2B-Base is the pre-trained text-only checkpoint, used to create all the LFM2.5-24B-A2B variants.

• The checkpoint is recommended for tasks requiring heavy fine-tuning. ^[1]

  This pre-trained checkpoint is only recommended for tasks that require heavy fine-tuning, like language-specific (e.g., Japanese) or domain-specific (e.g., medical) assistants, training on proprietary data, or experimenting with novel post-training approaches.

• An open model for tool-using agents, plus a checkpoint to fine-tune Daily brief 2026-10-13