LFM2.5-24B-A2B
LFM2.5-24B-A2B is a 24B-parameter reasoning model with only 2B active parameters, built for fast, private agents on-device. It is the largest and final model in the LFM2.5 family, which now spans two orders of magnitude: from LFM2.5-230M to LFM2.5-24B-A2B.
Params
24.0 B
Context
131,072
Downloads 30d
18 K
Likes
575
Download history
tracking started — chart appears after 7 days of snapshots (6 recorded)18 K downloads in the last 30 days
Can you run it?
Estimated VRAM at 8K context unless noted. Pick your hardware to see the verdict per quantization.
| File | Quant | Size | Est. VRAM | Verdict on RTX 4090 · 24 GB |
|---|---|---|---|---|
| model.safetensors | bf16 | 47.9 GB | 56.8 GB | ❌ Won’t fit |
| model.safetensors (bf16, full) | bf16 + 128K ctx | 47.9 GB | 110.8 GB | ❌ Won’t fit |
Estimate: file size × 1.1 + KV cache at 8K + 0.5 GB overhead. Not a benchmark — how we calculate this.
Run it
copy-paste, exact tags checked against the Hub$ curl -s https://aimodelscomparison.com/api/v1/models/lfm2-5-24b-a2b
{
"hf_id": "LiquidAI/LFM2.5-24B-A2B",
"params_b": 23.97,
"context_length": 131072,
"license": { "id": "other", "commercial": "unknown" },
"downloads_30d": 18100,
"vram_estimates": [
{ "quant": "bf16", "gb": 56.8 }
],
"updated_at": "2026-10-19T06:02:09Z"
}
Use via API
via OpenRouter · USD per 1M tokens · synced 2026-10-24$0.05
input / 1M
$0.70
output / 1M
$0.025
cached input / 1M
$0.75
1M in + 1M out
- Context window
- 128,000 tokens
- Max output
- 8,192 tokens
- Providers
- 1
- Cheapest provider
- Liquid · $0.05 / $0.70
- Best throughput
- —
- AA Intelligence / Coding / Agentic
- — / — / —
✓ Tool calling
✓ Reasoning
✓ Structured outputs
✗ Image input
✗ File input
✗ Audio input
✓ Prompt caching
Specifications
- Architecture
- Lfm2MoeForCausalLM
- Parameters
- 24.0 B
- Tensor type
- BF16
- Context length
- 131,072
- Vocabulary
- 128,000
- Layers / heads
- 40 / 32
- Licence
- other
- First seen on the Hub
- 2026-10-12
- Training datasets
- undisclosed
- Added to our catalog
- 2026-10-20
Family
Base model and the most-downloaded derivatives in the catalog.
Compare with any text-generation model