# LFM2.5-24B-A2B

Model tracker · checked 2026-10-27 · https://dbggr.dev/models/lfm2-5-24b-a2b/

Open-weight 24B MoE reasoning model (2B active) for fast, private agents on-device

LFM2.5-24B-A2B is a fast open-weight model from Liquid AI, released 13 October 2026. At Liquid AI it costs $0.05 per million input tokens and $0.70 per million output tokens, with a 131K-token context window.

- Made by: Liquid AI
- Released: 2026-10-13
- Status: Generally available, new
- Tier: Fast
- Weights: Open
- Memory to run: About 58 GB (BF16)
- Context window: 131,072 tokens
- Input: text
- Docs: https://docs.liquid.ai/lfm/models/lfm25-24b-a2b

## LFM2.5-24B-A2B price at Liquid AI

US dollars per million tokens. Blended is 3 input tokens for every output token, one number to compare on.

| Provider | Input | Output | Cached input | Blended | Batch | Checked |
|---|---|---|---|---|---|---|
| Liquid AI | $0.05 | $0.70 | $0.025 | $0.212 | – | [2026-10-27](https://www.liquid.ai/pricing) |

- Liquid AI: Liquid API. No batch price listed.

## Can you run LFM2.5-24B-A2B yourself?

Yes. Liquid AI publishes the weights on Hugging Face as [LiquidAI/LFM2.5-24B-A2B](https://huggingface.co/LiquidAI/LFM2.5-24B-A2B), under LFM Open License v1.0: 24B (MoE, 2B active) parameters, 48 GB in BF16.

Running it takes about 58 GB of GPU memory: one 80 GB data-centre GPU, or a Mac with comfortably more unified memory than that. That's the weight files plus 20% for working memory with short inputs; long contexts need more, and fitting in memory doesn't make it fast.

Liquid AI also publishes a smaller FP8 version, [LiquidAI/LFM2.5-24B-A2B-FP8](https://huggingface.co/LiquidAI/LFM2.5-24B-A2B-FP8), at 25 GB: about 30 GB to run, so a single GPU with at least 30 GB of memory, or a Mac with comfortably more than 30 GB.

## Models like LFM2.5-24B-A2B

| Model | Input | Output | Context | Weights |
|---|---|---|---|---|
| LFM2.5-24B-A2B | $0.05 | $0.70 | 131K | Open |
| [Gemma 4 31B](https://dbggr.dev/models/gemma-4-31b/) | $0.14 | $0.40 | 256K | Open |
| [Claude Haiku 5.5](https://dbggr.dev/models/claude-haiku-5-5/) | $0.10 | $0.50 | 1M | Closed |
| [GPT-6 Luna](https://dbggr.dev/models/gpt-6-luna/) | $0.10 | $0.50 | 1.1M | Closed |
| [Ministral 3 14B](https://dbggr.dev/models/ministral-3-14b/) | $0.20 | $0.20 | 256K | Open |
| [Qwen3.8 Flash](https://dbggr.dev/models/qwen3-8-flash/) | $0.15 | $0.47 | 1M | Closed |

## FAQ

### How much does LFM2.5-24B-A2B cost?

LFM2.5-24B-A2B costs $0.05 per million input tokens and $0.70 per million output tokens at Liquid AI, and $0.025 per million cached input tokens. Prices are list prices in US dollars, last checked 27 October 2026.

### What is LFM2.5-24B-A2B's context window?

LFM2.5-24B-A2B takes up to 131,072 tokens in one request.

### Is LFM2.5-24B-A2B open weight?

Yes. Liquid AI publishes the weights, so you can run LFM2.5-24B-A2B yourself or at any host that serves it.

### Can I run LFM2.5-24B-A2B locally?

LFM2.5-24B-A2B's weights are 48 GB in BF16, so it needs about 58 GB of GPU memory: one 80 GB data-centre GPU, or a Mac with comfortably more unified memory than that. Liquid AI also publishes a FP8 version that needs about 30 GB.

### Where can I use LFM2.5-24B-A2B?

LFM2.5-24B-A2B is available from Liquid AI.