back to local model checker

Hy3

MoEToolsThinkingPopular

Tencent · 295B (21B active) · Mixture of Experts

Production-focused agentic MoE with strong coding, tool use and long-context reasoning

Apache 2.0
Downloads

0

Likes

0

Released

2026-07

Context

256K tokens

Use cases

chatreasoningcode
Total experts192
Active experts8
Active params21.0B

Your hardware

16 GB Reference Desktop

Quant
Q4_K_M
Memory
151.6 GB
Speed
--
Fit
Too heavy

Quantization options

What is quantization?
QuantBitsVRAMDiskQualityStatus
Q2_K294.9 GB90.1 GBlowToo heavy · ~1 t/s
Q3_K_M3132.7 GB126.2 GBmoderateToo heavy · ~1 t/s
Q4_K_M4151.6 GB144.2 GBgoodToo heavy · ~1 t/s
Q5_K_M5189.4 GB180.3 GBgoodToo heavy · ~1 t/s
Q6_K6227.2 GB216.4 GBexcellentToo heavy · ~1 t/s
Q8_08302.7 GB288.5 GBexcellentToo heavy · ~1 t/s
F1616604.9 GB577 GBlosslessToo heavy · ~1 t/s

About this model

Hy3 (hy3) is a 295B sparse mixture-of-experts (MoE) model activating 21B active per token from Tencent in the Hunyuan family, designed primarily for chat, reasoning, code 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 256K context tokens; start at 4K–8K with Q8_0 KV cache on tight GPUs.

Hardware planning baselines start at 164.8 GB minimum memory and 274.7 GB recommended memory. At Q4_K_M, weights plus baseline runtime overhead use roughly 151.6 GB of RAM/VRAM and 144.2 GB of disk space. At 295B, Hy3 belongs firmly in the workstation and multi-GPU tier. Running it comfortably requires 274.7 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 Hy3 locally via llama.cpp with Flash Attention enabled. Under Apache License 2.0, review official weight cards before commercial redistribution.

Highlights

  • Local Fit Baseline164.8 GB minimum and 274.7 GB recommended RAM/VRAM.
  • Primary QuantQ4_K_M targets 151.6 GB memory and 144.2 GB disk.
  • Context & Capabilities256K max context · Sparse MoE · Tool calling · Reasoning.

System requirements by device

Memory planning targets 164.8 GB minimum RAM/VRAM and 274.7 GB recommended memory. The baseline quant is Q4_K_M, estimated at 151.6 GB memory and 144.2 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 151.6 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 (164.8 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_K94.9 GB90.1 GBlow
Q3_K_M132.7 GB126.2 GBmoderate
Q4_K_M151.6 GB144.2 GBgood
Q5_K_M189.4 GB180.3 GBgood
Q6_K227.2 GB216.4 GBexcellent
Q8_0302.7 GB288.5 GBexcellent
F16604.9 GB577.0 GBlossless

How to run Hy3 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 Hy3 on Multi-GPU or High-RAM Desktop

  1. Plan for 164.8 GB minimum combined memory and 274.7 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 Hy3 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 151.6 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

Hy3 vs alternatives

ModelParamsContextMin RAMLicense
Hy3295B (21B active)256K164.8 GBApache 2.0
HunyuanImage 3.080B (13B active)4K44.7 GBTencent Hunyuan Community
HunyuanImage 3.0 Instruct80B (13B active)4K44.7 GBTencent Hunyuan Community
HunyuanVideo 1.58.3B4K4.6 GBTencent Hunyuan Community
Qwen 3 235B-A22B235B (22B active)128K131.3 GBApache 2.0
Qwen 3 VL 235B-A22B235B (22B active)256K131.3 GBApache 2.0

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 Hy3?
Hy3 is a 295B sparse mixture-of-experts model (21B active active per token) from Tencent, supporting up to 256K tokens with 164.8 GB minimum RAM and 274.7 GB recommended RAM.
How do I download and install hy3?
Download a GGUF build for Hy3, start with Q4_K_M, and confirm the model fits locally before increasing context or quant quality.
How much VRAM or RAM does Hy3 require?
At Q4_K_M, Hy3 occupies roughly 151.6 GB of active memory and 144.2 GB on disk. Add 2–6 GB of spare headroom for KV cache buffers, runtime scratch space, and OS apps (274.7 GB recommended).
What is the best way to run Hy3 locally?
Run Hy3 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 Hy3 run on a consumer laptop or smartphone?
No smartphone or 8–16 GB laptop can hold Hy3 cleanly in local memory. It targets multi-GPU NVIDIA workstations, high-RAM Mac Studios, or MoE CPU-RAM hybrid rigs with at least 164.8 GB total memory.
How does Hy3 compare to HunyuanImage 3.0?
HunyuanImage 3.0 (80B, 44.7 GB min RAM) sits closest in size. Compare active parameters, context ceiling (256K vs 4K), and license terms (Apache 2.0) before switching.