meituan-longcat/LongCat-2.5
main boardmodel
LongCat-2.5 is a native multimodal MoE model for long-horizon agentic tasks, with 1.6 trillion total parameters and approximately 48 billion activated per token. Mixture-of-experts, 1780B total (768 experts). Context: 512K tokens (config). License: mit.
- Org
- meituan-longcat
- Released
- 2026-11-04
- Status
- main board
- License
- mit (permissive)
- Parameters
- 1779.8B
- Architecture
- MoE · 768 experts, None/token · LongcatCausalLM
- Context
- 524,288 tokens (config.json)
- Task
- image-text-to-text · vision
- Signal
- 268 HF likes · 1573 downloads
Runs on
mixed: real files + estimate; need = weights + FP16 KV cache for 8K tokens + 1GB. A sizing calculation, not tested.
- Q4 ~1220.2 GB (estimated size): no listed device
- Q8 ~2043.3 GB (estimated size): no listed device
- FP16 ~3714.3 GB: no listed device
Links
First published here 2026-11-04 · updated 2026-11-10 · markdown · JSON