Harness 1 — Hardware Requirements & GPU Compatibility
Chat
Harness 1 is a 20.9B-parameter open language model from pat-jj. It supports a context window of up to 131,072 tokens. At Q4_K_M it needs about 12.92 GB of VRAM — see which GPUs and Macs can run it below.
Runs great on a 24 GB GPU
Runs from 12 GB (Q3_K_M · 64K)Best on 24 GB (Q4_K_M · ~190 tok/s · 128K)Macs from 16 GB (Q3_K_M · ~50 tok/s)
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run Harness 1
One recommended way. Every other method, OS and quant is in the step-by-step guide.
Recommended
vllm serve pat-jj/harness-1 --max-model-len 16384vLLM repo with its own quantisation
Download: Weights on Hugging Face
Other ways
each opens step-by-stepRun in cloud
Fits on RTX 5060 Ti 16GB (2 GB headroom) · Q4_K_M
- Generation speed
- ~98 tok/s
- generation speed
- Cost per 1M output tokens
- $0.47
- per 1M output tokens
How Much VRAM Does Harness 1 Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 9.3 GB | 9.8 GB | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Fits: 128K context · ~241 tok/s | |
| 10.6 GB | 11.1 GB | 3-bit medium — 3-bit medium quantization | Fits: 128K context · ~220 tok/s | |
| 12.9 GB | 13.4 GB | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Fits: 128K context · ~190 tok/s | |
| 15.3 GB | 15.8 GB | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Fits: 128K context · ~167 tok/s | |
| 17.6 GB | 18.1 GB | 6-bit, very good — 6-bit quantization, very good quality | Fits: 128K context · ~149 tok/s | |
| 21.3 GB | 21.8 GB | 8-bit, near-lossless — 8-bit quantization, near-lossless | Fits: 128K context · ~128 tok/s | |
| 42.2 GB | 42.7 GB | 16-bit · bfloat — Brain floating point 16 — preferred for training | Does not fit: needs 42.4 GB |
Selected Q4_K_M
4-bit medium quantization13.1GBat 16K context
- Weights
- 12.6 GB
- KV cache
- 0.3 GB
- Overhead
- 0.3 GB
Fits RTX 4090 (reference) · 10.9 GB free · ~190 tok/s
≥50 tok/s needs ≥213 GB/s · RTX 5090 ~282, RTX 3090 Ti ~190
Which Hardware Can Run Harness 1?
Q4_K_M · 12.9 GBHarness 1 (Q4_K_M) requires 12.9 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 17+ GB is recommended. Using the full 131K context window can add up to 2.2 GB, bringing total usage to 15.1 GB. 26 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti, NVIDIA GeForce RTX 5080. 125 devices with unified memory can run Harness 1, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), MacBook Pro 14" M3 Pro (18 GB).
16–18 GB
16–18 GB tier →fits · 2.9 GB free · up to 128K ctx
18 GPUs · 1 Mac class · 1 device family
RTX 5080~183Simulate output on RTX 5080RTX 5070 Ti~173Simulate output on RTX 5070 TiRTX 4080 SUPER~149Simulate output on RTX 4080 SUPERM3 Pro*18 GB~34Simulate output on MacBook Pro 14" M3 Pro (18 GB)Orin NX~25Simulate output on NVIDIA Jetson Orin NX 16GB20 GB
fits · 6.9 GB free · up to 128K ctx
1 GPU
RX 7900 XT~98Simulate output on RX 7900 XT32–40 GB
fits · 18.9 GB free · up to 128K ctx
1 GPU · 3 Mac classes · 4 device families
RTX 5090~282Simulate output on RTX 5090M5 Max36 GB~78Simulate output on Mac Studio M5 Max (36 GB)M532 GB~35Simulate output on MacBook Air 13" M5 (32 GB)Orin~48Simulate output on NVIDIA Jetson AGX Orin 32GBSnapdragon X Elite (X1E-80-100)~31Simulate output on Snapdragon X Elite Copilot+ PC
By VRAM tier · typical GPUs and Macs at Q4_K_M
Similar models
4| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| InternVL3 5 GPT OSS 20B A4B Preview HFOpenGVLab | — | 13.6 GB | ~180 | same size |
| Internlm 20BInternLM | — | 45.3 GB | —needs > 24 GB | same size |
| ERNIE 4.5 21B A3B ThinkingBaidu | — | 14.3 GB | ~213 | same size |
| Mistral Small Instruct 2409Mistral AI | — | 17.4 GB | ~46 | same size |
Specifications
- Publisher
- pat-jj
- Parameters
- 20.9B
- Active parameters
- 4.2B per token
- Architecture
- GptOssForCausalLM
- Context Length
- 131,072 tokens
- Vocabulary Size
- 201,088
- Release Date
- 2026-05-12
Frequently Asked Questions
How much VRAM does Harness 1 need?
Harness 1 requires 12.9 GB of VRAM at Q4_K_M, or 42.2 GB at BF16. Full 131K context adds up to 2.2 GB (15.1 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 20.9B × 4.8 bits ÷ 8 = 12.5 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 ≈ 2.6 GB (at full 131K context)
VRAM usage by quantization
Can NVIDIA GeForce RTX 4090 run Harness 1?
Yes, at Q8_0 (21.3 GB) or lower. Higher quantizations like BF16 (42.2 GB) exceed the NVIDIA GeForce RTX 4090's 24 GB.
What's the best quantization for Harness 1?
For Harness 1, Q4_K_M (12.9 GB) offers the best balance of quality and VRAM usage. Q5_K_M (15.3 GB) provides better quality if you have the VRAM. The smallest option is Q2_K at 9.3 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run Harness 1 on a Mac?
Yes — Mac Mini M2 Pro (16 GB) and 133 other Macs can run Harness 1, the smallest at Q3_K_M. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 12.9 GB of usable unified memory (RAM minus macOS overhead), from MacBook Pro 14" M3 Pro (18 GB).
Can I run Harness 1 locally?
Yes — Harness 1 can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Harness 1?
At Q4_K_M, Harness 1 can reach ~187 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~190 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 → 2.6 GB active ÷ (8000 × 0.65) = 0.498 ms, plus 24 layers × 0.055 ms = 1.320 ms, so 1000 ÷ 1.818 ms = ~550 tok/s
Estimated speed at Q4_K_M (12.9 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of Harness 1?
At Q4_K_M, the download is about 12.55 GB. The full-precision BF16 version is 41.83 GB. The smallest option (Q2_K) is 8.89 GB.
Which GPUs can run Harness 1?
26 consumer GPUs can run Harness 1 at Q4_K_M. Top options include AMD Radeon RX 7900 XT, AMD Radeon RX 7900 XTX, NVIDIA GeForce RTX 3090, AMD Radeon RX 6800. 8 GPUs have plenty of headroom for comfortable inference.
Which devices can run Harness 1?
125 devices with unified memory can run Harness 1 at Q4_K_M (12.9 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.