back to local model checker

LongCat 2.0

MoEToolsThinkingPopular

Meituan · 1.6T (48B active) · Mixture of Experts

Frontier-scale agentic and coding MoE with sparse attention and native 1M context

MIT
Downloads

0

Likes

0

Released

2026-07

Context

1M tokens

Use cases

chatcodereasoning

Your hardware

16 GB Reference Desktop

Quant
Q4_K_M
Memory
820.1 GB
Speed
--
Fit
Too heavy

Quantization options

What is quantization?
QuantBitsVRAMDiskQualityStatus
Q2_K2512.7 GB488.9 GBlowToo heavy · ~1 t/s
Q3_K_M3717.6 GB684.5 GBmoderateToo heavy · ~1 t/s
Q4_K_M4820.1 GB782.3 GBgoodToo heavy · ~1 t/s
Q5_K_M51025 GB977.9 GBgoodToo heavy · ~1 t/s
Q6_K61229.8 GB1173.5 GBexcellentToo heavy · ~1 t/s
Q8_081639.6 GB1564.6 GBexcellentToo heavy · ~1 t/s
F16163278.8 GB3129.2 GBlosslessToo heavy · ~1 t/s

About this model

LongCat 2.0 (longcat 2.0) is a 1.6T sparse mixture-of-experts (MoE) model activating 48B active per token from Meituan in the LongCat family, designed primarily for chat, code, reasoning workflows. Native reasoning/thinking tokens expand output length per query, so reserve extra KV cache headroom. Supports structured tool calling; use Q4_K_M or higher when strict JSON schema adherence matters. Supports up to 1M context tokens; start at 4K–8K with Q8_0 KV cache on tight GPUs.

Hardware planning baselines start at 894.1 GB minimum memory and 1490.1 GB recommended memory. At Q4_K_M, weights plus baseline runtime overhead use roughly 820.1 GB of RAM/VRAM and 782.3 GB of disk space. At 1.6T, LongCat 2.0 belongs firmly in the workstation and multi-GPU tier. Running it comfortably requires 1490.1 GB of total memory across dual RTX GPUs, high-memory Apple Silicon (Mac Studio Ultra/Max), or MoE hybrid CPU-RAM offloading—mobile devices and standard 16 GB laptops are too constrained for practical speeds.

Deploy LongCat 2.0 locally via llama.cpp with Flash Attention enabled. Under MIT License, review official weight cards before commercial redistribution.

Highlights

  • Local Fit Baseline894.1 GB minimum and 1490.1 GB recommended RAM/VRAM.
  • Primary QuantQ4_K_M targets 820.1 GB memory and 782.3 GB disk.
  • Context & Capabilities1M max context · Sparse MoE · Tool calling · Reasoning.

System requirements by device

Memory planning targets 894.1 GB minimum RAM/VRAM and 1490.1 GB recommended memory. The baseline quant is Q4_K_M, estimated at 820.1 GB memory and 782.3 GB disk. Actual runtime usage shifts with context length, KV cache precision, and OS background overhead.

DeviceNotes
Mac (Apple silicon)Requires a 48 GB–192 GB Mac Studio or MacBook Pro Max/Ultra with enough unified wired budget for 820.1 GB.
Windows / Linux GPUUse dual/multi-GPU CUDA splitting (--tensor-split) or MoE CPU expert offloading (-ot) paired with 32–128 GB DDR5 system RAM.
Phones & Compact Edge BoardsNot practical on phones (894.1 GB min RAM exceeds Android LMK and iOS Jetsam limits). Pick a 2B–4B model on mobile.

Start with a short session and leave memory for the operating system. A file fitting on disk does not guarantee that it will load in memory.

Estimated memory by quantization

These are rough estimates, not tested minimum requirements. Check the actual file and runtime documentation; context, offloading and operating-system overhead can change the memory needed.

Quant TierEstimated memoryDisk SizeQuality
Q2_K512.7 GB488.9 GBlow
Q3_K_M717.6 GB684.5 GBmoderate
Q4_K_M820.1 GB782.3 GBgood
Q5_K_M1025.0 GB977.9 GBgood
Q6_K1229.8 GB1173.5 GBexcellent
Q8_01639.6 GB1564.6 GBexcellent
F163278.8 GB3129.2 GBlossless

How to run LongCat 2.0 on your device

Start with a short prompt and the smallest quant that still gives the quality you need. A clean first run is more useful than loading a larger file with no memory headroom.

Windows

Run LongCat 2.0 on Multi-GPU or High-RAM Desktop

  1. Plan for 894.1 GB minimum combined memory and 1490.1 GB recommended memory for Q4_K_M.
  2. Use llama.cpp `-ot ".ffn_.*_exps.=CPU"` or `--tensor-split` across multiple GPUs so attention and KV cache remain in fast VRAM.
  3. Enable Flash Attention (`-fa on`) and Q8_0 KV cache (`-ctk q8_0 -ctv q8_0`) to prevent context buffers from exceeding GPU headroom.
  4. Avoid attempting to run workstation-scale weights on smartphones or 8 GB laptops; use a 2B–8B checkpoint on portable devices.
Device how-to guide
Mac

Run LongCat 2.0 on High-Memory Apple Silicon Mac

  1. Verify you are on a 48 GB–192 GB Mac Studio or MacBook Pro Max/Ultra capable of holding 820.1 GB in wired unified memory.
  2. Launch via `llama-server` or LM Studio with MLX/Metal enabled and set initial context conservatively to 4K–8K.
  3. Monitor Activity Monitor Memory Pressure to confirm zero SSD swap thrashing during token generation.
  4. Step down to a lighter quantization (e.g., Q3_K_M or IQ4_XS) if total memory utilization exceeds 75%.
Device how-to guide

LongCat 2.0 vs alternatives

ModelParamsContextMin RAMLicense
LongCat 2.01.6T (48B active)1M894.1 GBMIT
DeepSeek V4 Pro1.6T (49B active)1M894.1 GBMIT
Kimi K2.61.06T (32B active)256K591.2 GBKimi
Kimi K21T (32B active)128K558.8 GBKimi
Qwen 3.8 2.4T-A95B2.4T (95B active)1M1341.1 GBQwen
GLM-5.1754B (40B active)128K421.3 GBMIT

Use this table to choose the smallest nearby model that still matches the job. A related model with lower RAM can be the better local choice when the current page fits only at the edge of available memory.

Frequently asked questions

What is LongCat 2.0?
LongCat 2.0 is a 1.6T sparse mixture-of-experts model (48B active active per token) from Meituan, supporting up to 1M tokens with 894.1 GB minimum RAM and 1490.1 GB recommended RAM.
How do I download and install longcat 2.0?
Download a GGUF build for LongCat 2.0, start with Q4_K_M, and confirm the model fits locally before increasing context or quant quality.
How much VRAM or RAM does LongCat 2.0 require?
At Q4_K_M, LongCat 2.0 occupies roughly 820.1 GB of active memory and 782.3 GB on disk. Add 2–6 GB of spare headroom for KV cache buffers, runtime scratch space, and OS apps (1490.1 GB recommended).
What is the best way to run LongCat 2.0 locally?
Run LongCat 2.0 in llama.cpp starting with Q4_K_M. Keep initial context at 4K–8K tokens and confirm 100% GPU/Metal offload before scaling context length.
Can LongCat 2.0 run on a consumer laptop or smartphone?
No smartphone or 8–16 GB laptop can hold LongCat 2.0 cleanly in local memory. It targets multi-GPU NVIDIA workstations, high-RAM Mac Studios, or MoE CPU-RAM hybrid rigs with at least 894.1 GB total memory.
How does LongCat 2.0 compare to DeepSeek V4 Pro?
DeepSeek V4 Pro (1.6T, 894.1 GB min RAM) sits closest in size. Compare active parameters, context ceiling (1M vs 1M), and license terms (MIT) before switching.