Hy3 REAP 200B 21B — Hardware Requirements & GPU Compatibility
Chat
Hy3 REAP 200B 21B is a 199.6B-parameter open language model from OpenMOSE in the Hunyuan 3 family. It supports a context window of up to 262,144 tokens. At Q4_K_M it needs about 120.39 GB of VRAM — see which GPUs and Macs can run it below.
Hy3 REAP 200B 21B runs on a 128 GB+ Mac or a datacenter GPU
Best local pick → Mac Pro M2 Ultra (128 GB) · Q4_K_M · 122.7 GB · ~28 tok/s
Macs from 96 GB (Q2_K · ~34 tok/s)
Apache 2.0
Not set up for it?
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run Hy3 REAP 200B 21B
One recommended way. Every other method, OS and quant is in the step-by-step guide.
Recommended
ollama run hf.co/OpenMOSE/Hy3-REAP-200B-21B-GGUF:Q4_K_MQ4_K_M runs on one Instinct MI250X 128 GB · 16K context
Download: Q4_K_M/Hy3-REAP-200B-GGUF-Q4_K_M-00001-of-00007.gguf122.40 GB + 6 partsOpenMOSE
Other ways
each opens step-by-stepRun in cloud
Fits on B200 (192 GB) (69 GB headroom) · Q4_K_M
- Generation speed
- ~153 tok/s
- generation speed
- Cost per 1M output tokens
- $10.85
- per 1M output tokens
How Much VRAM Does Hy3 REAP 200B 21B Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | Runs on | |
|---|---|---|---|---|
| 68 GB | 73 GB | 2-bit · i-quant medium — Importance-weighted 2-bit, medium | Runs on RTX PRO 5000 Blackwell 72 GB | |
| 78.0 GB | 83 GB | 3-bit · i-quant — Importance-weighted 3-bit | Runs on RTX PRO 6000 Blackwell Max-Q Workstation Edition 96 GB | |
| 85.5 GB | 90.5 GB | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Runs on RTX PRO 6000 Blackwell Max-Q Workstation Edition 96 GB | |
| 97.9 GB | 103 GB | 3-bit medium — 3-bit medium quantization | Runs on Instinct MI250X 128 GB | |
| 107.9 GB | 112.9 GB | 4-bit · i-quant compact — Importance-weighted 4-bit, compact | Runs on Instinct MI250X 128 GB | |
| 120.4 GB | 125.4 GB | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Runs on Instinct MI250X 128 GB | |
| 142.8 GB | 147.9 GB | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Runs on B200 192 GB | |
| 165.3 GB | 170.3 GB | 6-bit, very good — 6-bit quantization, very good quality | Runs on B200 192 GB | |
| 200.2 GB | 205.3 GB | 8-bit, near-lossless — 8-bit quantization, near-lossless | Runs on Instinct MI325X 256 GB | |
| 399.8 GB | 404.8 GB | 16-bit · bfloat — Brain floating point 16 — preferred for training | Runs on 6× H100 SXM 80 GB |
Selected Q4_K_M
4-bit medium quantization122.7GBat 16K context
- Weights
- 119.8 GB
- KV cache
- 2.7 GB
- Overhead
- 0.3 GB
Server-class: ~122.7 GB at 16K context · one Instinct MI250X 128 GB
Which Hardware Can Run Hy3 REAP 200B 21B?
Q4_K_M · 120.4 GBHy3 REAP 200B 21B (Q4_K_M) requires 120.4 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 157+ GB is recommended. Using the full 262K context window can add up to 42.6 GB, bringing total usage to 163 GB. No consumer GPU has enough memory. 15 devices with unified memory can run Hy3 REAP 200B 21B, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Pro M2 Ultra (128 GB).
96 GB
Q4_K_M · best fit Q2_K (87.8 GB)
3 GPUs · 2 Mac classes
RTX PRO 6000 Blackwell Max-Q Workstation Edition~90Simulate output on RTX PRO 6000 Blackwell Max-Q Workstation EditionRTX PRO 6000 Blackwell Workstation Edition~90Simulate output on RTX PRO 6000 Blackwell Workstation EditionRTX PRO 6000 Blackwell Server Edition~83Simulate output on RTX PRO 6000 Blackwell Server EditionM3 Ultra*96 GB~34Simulate output on Mac Studio (M3 Ultra, 96GB)M3 Max*96 GB~19Simulate output on MacBook Pro 14" M3 Max (96 GB)128–144 GB
fits · 5.3 GB free · up to 32K ctx
4 GPUs · 2 Mac classes
GH200 Grace Hopper Superchip~127Simulate output on GH200 Grace Hopper SuperchipH200 NVL~126Simulate output on H200 NVLH200 SXM~126Simulate output on H200 SXMM2 Ultra*128 GB~28Simulate output on Mac Pro M2 Ultra (128 GB)M5 Max128 GB~24Simulate output on Mac Studio M5 Max (128 GB)192+ GB
fits · 69.3 GB free · up to 256K ctx
6 GPUs · 1 Mac class · 2 device families
B200~153Simulate output on B200B300~153Simulate output on B300Instinct MI350X~55Simulate output on Instinct MI350XM3 Ultra*256–512 GB~28Simulate output on Mac Studio (M3 Ultra, 256GB)8x H100 SXM5~199Simulate output on NVIDIA DGX H1008x A100 SXM4~184Simulate output on NVIDIA DGX A100 640GB
By VRAM tier · typical GPUs and Macs at Q4_K_M
Consumer GPUs (up to 32 GB) and Macs up to 64 GB: too small for this model (needs ≥ 70.4 GB)8 GB tier →12 GB tier →16 GB tier →24 GB tier →Macs →
Similar models
1| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| Step 3.7 FlashStepFun | — | 139.2 GB | —needs > 24 GB | same size |
Specifications
- Publisher
- OpenMOSE
- Family
- Hunyuan 3
- Parameters
- 199.6B
- Active parameters
- 18.4B per token
- Architecture
- HYV3ForCausalLM
- Context Length
- 262,144 tokens
- Vocabulary Size
- 120,832
- Release Date
- 2026-07-16
- License
- Apache 2.0
Frequently Asked Questions
How much VRAM does Hy3 REAP 200B 21B need?
Hy3 REAP 200B 21B requires 120.4 GB of VRAM at Q4_K_M, or 399.8 GB at BF16. Full 262K context adds up to 42.6 GB (163 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 199.6B × 4.8 bits ÷ 8 = 119.8 GB
KV Cache + Overhead ≈ 0.6 GB (at 2K context + ~0.3 GB framework)
Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.
KV Cache + Overhead ≈ 43.2 GB (at full 262K context)
VRAM usage by quantization
Can NVIDIA GeForce RTX 5090 run Hy3 REAP 200B 21B?
No — Hy3 REAP 200B 21B requires at least 68 GB at IQ2_M, which exceeds the NVIDIA GeForce RTX 5090's 32 GB of VRAM.
What's the best quantization for Hy3 REAP 200B 21B?
For Hy3 REAP 200B 21B, Q4_K_M (120.4 GB) offers the best balance of quality and VRAM usage. Q5_K_M (142.8 GB) provides better quality if you have the VRAM. The smallest option is IQ2_M at 68 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run Hy3 REAP 200B 21B on a Mac?
Yes — Mac Studio (M3 Ultra, 96GB) and 20 other Macs can run Hy3 REAP 200B 21B, the smallest at Q2_K. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 120.4 GB of usable unified memory (RAM minus macOS overhead), from Mac Pro M2 Ultra (128 GB).
Can I run Hy3 REAP 200B 21B locally?
Yes — Hy3 REAP 200B 21B can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Hy3 REAP 200B 21B?
At Q4_K_M, Hy3 REAP 200B 21B can reach ~55 tok/s on AMD Instinct MI350X. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = 1000 ÷ (active GB ÷ (bandwidth GB/s × efficiency) × 1000 + layers × routing ms)
Mixture-of-Experts: just the active experts get read for each token, along with a fixed routing cost per layer.
Example: NVIDIA B200 → 11.1 GB active ÷ (8000 × 0.65) = 2.13 ms, plus 80 layers × 0.055 ms = 4.40 ms, so 1000 ÷ 6.53 ms = ~153 tok/s
Estimated speed at Q4_K_M (120.4 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of Hy3 REAP 200B 21B?
At Q4_K_M, the download is about 119.75 GB. The full-precision BF16 version is 399.17 GB. The smallest option (IQ2_M) is 67.36 GB.
Which GPUs can run Hy3 REAP 200B 21B?
No single consumer GPU has enough VRAM to run Hy3 REAP 200B 21B at Q4_K_M (120.4 GB). Multi-GPU or professional hardware is required.
Which devices can run Hy3 REAP 200B 21B?
15 devices with unified memory can run Hy3 REAP 200B 21B at Q4_K_M (120.4 GB), including Mac Pro M2 Ultra (128 GB), Mac Pro M2 Ultra (192 GB), Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.