Qwen3 Coder REAP 25B A3B — Hardware Requirements & GPU Compatibility
ChatCode
Qwen3 Coder REAP 25B A3B is a 24.9B-parameter open language model from Cerebras in the Qwen 3 family. It supports a context window of up to 262,144 tokens. At Q4_K_M it needs about 15.32 GB of VRAM — see which GPUs and Macs can run it below.
Runs great on a 24 GB GPU
Runs from 12 GB (Q2_K · 16K)Best on 24 GB (Q4_K_M · ~173 tok/s · 128K)Macs from 16 GB (Q2_K · ~50 tok/s)
Apache 2.0
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run Qwen3 Coder REAP 25B A3B
One recommended way. Every other method, OS and quant is in the step-by-step guide.Each way opens a step-by-step guide for your OS and quant.
Recommended
vllm serve cerebras/Qwen3-Coder-REAP-25B-A3B --max-model-len 16384vLLM repo with its own quantisation
Download: Weights on Hugging Face
Ways to runOther ways
each opens step-by-stepRun in cloud
Runs on RTX 3090 (24 GB) (8 GB headroom) · Q4_K_M
- Generation speed
- ~166 tok/s
- generation speed
- Cost per 1M output tokens
- $0.28
- per 1M output tokens
How Much VRAM Does Qwen3 Coder REAP 25B A3B Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 11.0 GB | 12.5 GB | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Fits: 256K context · ~204 tok/s | |
| 12.5 GB | 14 GB | 3-bit medium — 3-bit medium quantization | Fits: 128K context · ~192 tok/s | |
| 15.3 GB | 16.8 GB | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Fits: 128K context · ~173 tok/s | |
| 18.1 GB | 19.6 GB | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Fits: 64K context · ~157 tok/s | |
| 20.9 GB | 22.4 GB | 6-bit, very good — 6-bit quantization, very good quality | Fits: 32K context · ~144 tok/s | |
| 25.3 GB | 26.8 GB | 8-bit, near-lossless — 8-bit quantization, near-lossless | Does not fit: needs 26 GB | |
| 50.1 GB | 51.7 GB | 16-bit · bfloat — Brain floating point 16 — preferred for training | Does not fit: needs 50.8 GB |
Selected Q4_K_M
4-bit medium quantization16.0GBat 16K context
- Weights
- 14.9 GB
- KV cache
- 0.8 GB
- Overhead
- 0.3 GB
Fits RTX 4090 (reference) · 8.0 GB free · ~173 tok/s
≥50 tok/s needs ≥183 GB/s · RTX 5090 ~227, RTX 3090 Ti ~173
Which Hardware Can Run Qwen3 Coder REAP 25B A3B?
Q4_K_M · 15.3 GBQwen3 Coder REAP 25B A3B (Q4_K_M) requires 15.3 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 20+ GB is recommended. Using the full 262K context window can add up to 12.8 GB, bringing total usage to 28.1 GB. 26 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti, NVIDIA GeForce RTX 5080. 122 devices with unified memory can run Qwen3 Coder REAP 25B A3B, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Mini M5 Pro (24 GB).
20 GB
fits · 4.0 GB free · up to 64K ctx
1 GPU
RX 7900 XT~72Simulate output on RX 7900 XT32–40 GB
fits · 16.0 GB free · up to 256K ctx
1 GPU · 3 Mac classes · 4 device families
RTX 5090~227Simulate output on RTX 5090M5 Max36 GB~63Simulate output on Mac Studio M5 Max (36 GB)M532 GB~35Simulate output on MacBook Air 13" M5 (32 GB)Orin~55Simulate output on NVIDIA Jetson AGX Orin 32GBSnapdragon X Elite (X1E-80-100)~35Simulate output on Snapdragon X Elite Copilot+ PC
By VRAM tier · typical GPUs and Macs at Q4_K_M
Alternatives to Qwen3 Coder REAP 25B A3B
Qwen 3 family · 79 models →Similar models
6| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| LFM2.5 24B A2BLiquid AI | — | 15 GB | ~221 | same size |
| Diffusiongemma 26B A4B ITGoogle | — | 16.5 GB | ~171 | same size |
| Gemma 4 26B A4BGoogle | — | 16.5 GB | ~171 | same size |
| Devstral Small 2 24B Instruct 2512Mistral AI | — | 18.1 GB | ~43 | same size |
| Ariarhymes-ai | — | 20.2 GB | ~168 | same size |
| Trinity MiniArcee AI | — | 16.5 GB | ~199 | same size |
Specifications
- Publisher
- Cerebras
- Family
- Qwen 3
- Parameters
- 24.9B
- Active parameters
- 3.4B per token
- Architecture
- Qwen3MoeForCausalLM
- Context Length
- 262,144 tokens
- Vocabulary Size
- 151,936
- Release Date
- 2025-10-20
- License
- Apache 2.0
Frequently Asked Questions
How much VRAM does Qwen3 Coder REAP 25B A3B need?
Qwen3 Coder REAP 25B A3B requires 15.3 GB of VRAM at Q4_K_M, or 50.1 GB at BF16. Full 262K context adds up to 12.8 GB (28.1 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 24.9B × 4.8 bits ÷ 8 = 14.9 GB
KV Cache + Overhead ≈ 0.4 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 ≈ 13.2 GB (at full 262K context)
VRAM usage by quantization
Can NVIDIA GeForce RTX 4090 run Qwen3 Coder REAP 25B A3B?
Yes, at Q6_K (20.9 GB) or lower. Higher quantizations like Q8_0 (25.3 GB) exceed the NVIDIA GeForce RTX 4090's 24 GB.
What's the best quantization for Qwen3 Coder REAP 25B A3B?
For Qwen3 Coder REAP 25B A3B, Q4_K_M (15.3 GB) offers the best balance of quality and VRAM usage. Q5_K_M (18.1 GB) provides better quality if you have the VRAM. The smallest option is Q2_K at 11.0 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run Qwen3 Coder REAP 25B A3B on a Mac?
Yes — Mac Mini M2 Pro (16 GB) and 133 other Macs can run Qwen3 Coder REAP 25B A3B, 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 15.3 GB of usable unified memory (RAM minus macOS overhead), from Mac Mini M5 Pro (24 GB).
Can I run Qwen3 Coder REAP 25B A3B locally?
Yes — Qwen3 Coder REAP 25B A3B can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Qwen3 Coder REAP 25B A3B?
At Q4_K_M, Qwen3 Coder REAP 25B A3B can reach ~100 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~173 tok/s. 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 → 2.1 GB active ÷ (8000 × 0.65) = 0.40 ms, plus 48 layers × 0.055 ms = 2.64 ms, so 1000 ÷ 3.04 ms = ~329 tok/s
Estimated speed at Q4_K_M (15.3 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of Qwen3 Coder REAP 25B A3B?
At Q4_K_M, the download is about 14.92 GB. The full-precision BF16 version is 49.73 GB. The smallest option (Q2_K) is 10.57 GB.
Which GPUs can run Qwen3 Coder REAP 25B A3B?
26 consumer GPUs can run Qwen3 Coder REAP 25B A3B at Q4_K_M. Top options include AMD Radeon RX 7900 XTX, NVIDIA GeForce RTX 3090, NVIDIA GeForce RTX 3090 Ti, AMD Radeon RX 6800. 7 GPUs have plenty of headroom for comfortable inference.
Which devices can run Qwen3 Coder REAP 25B A3B?
122 devices with unified memory can run Qwen3 Coder REAP 25B A3B at Q4_K_M (15.3 GB), including AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.