ZAYA1 VL 8B — Hardware Requirements & GPU Compatibility
VisionZAYA1-VL-8B serves as Zyphra's vision-language model with roughly 8 billion total parameters. It builds on the ZAYA1-8B language model, using a Qwen2.5-VL vision encoder. The card outlines two key changes: LoRA parameters specific to vision that activate only when image tokens are present, and bidirectional attention across those same tokens. According to the card, it contains roughly 0.7 billion active parameters, was trained exclusively on open data, and shows strong performance for its scale—scoring 92.5 on DocVQA and 87.5 on AI2D. At this size, quantization allows the model to fit on a single consumer GPU. The context window spans 32,768 tokens. Released under the Apache 2.0 license, it permits unrestricted commercial and research use. Published in April 2026, this release extends the ZAYA1 base text model to handle image understanding tasks.
Runs great on a 24 GB GPU
Runs from 8 GB (Q4_K_M · 16K)Best on 24 GB (Q4_K_M · ~346 tok/s · 32K)Macs from 16 GB (Q4_K_M · ~89 tok/s)
Apache 2.0
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run ZAYA1 VL 8B
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
ollama run hf.co/1bit-MONSTER/ZAYA1-VL-8B-GGUF:Q4_K_MQ4_K_M fits fully on the RTX 4090 · 16.5 GB free · 16K context
Download: ZAYA1-VL-8B-Q4_K_M.gguf5.71 GB1bit-MONSTER
Ways to runOther ways
each opens step-by-stepRun in cloud
Your own private copy · Q4_K_M · 24 GB GPU
- The easiest way to try this model
- Running in a few minutes, easy setup
- Chat right here or connect any app
$5+ bonus for new users
Runs on RTX 3060 12GB (4 GB headroom) · Q4_K_M
- Generation speed
- ~242 tok/s
- generation speed
- Cost per 1M output tokens
- $0.073
- per 1M output tokens
How Much VRAM Does ZAYA1 VL 8B Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 4.6 GB | 7.1 GB | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Fits: 32K context · ~370 tok/s | |
| 5.2 GB | 7.7 GB | 3-bit medium — 3-bit medium quantization | Fits: 32K context · ~361 tok/s | |
| 6.3 GB | 8.8 GB | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Fits: 32K context · ~346 tok/s | |
| 7.4 GB | 9.9 GB | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Fits: 32K context · ~332 tok/s | |
| 8.5 GB | 11 GB | 6-bit, very good — 6-bit quantization, very good quality | Fits: 32K context · ~319 tok/s | |
| 10.2 GB | 12.7 GB | 8-bit, near-lossless — 8-bit quantization, near-lossless | Fits: 32K context · ~301 tok/s | |
| 19.9 GB | 22.4 GB | 16-bit · bfloat — Brain floating point 16 — preferred for training | Fits: 32K context · ~228 tok/s |
Selected Q4_K_M
4-bit medium quantization7.5GBat 16K context
- Weights
- 5.8 GB
- KV cache
- 1.3 GB
- Overhead
- 0.3 GB
Fits RTX 4090 (reference) · 16.5 GB free · ~346 tok/s
≥50 tok/s needs ≥39 GB/s · RTX 5090 ~386, RTX 3090 Ti ~346
Which Hardware Can Run ZAYA1 VL 8B?
Q4_K_M · 6.3 GBZAYA1 VL 8B (Q4_K_M) requires 6.3 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 9+ GB is recommended. Using the full 33K context window can add up to 2.5 GB, bringing total usage to 8.8 GB. 95 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti, NVIDIA GeForce RTX 3070 Ti. 172 devices with unified memory can run ZAYA1 VL 8B, including NVIDIA RTX Spark N1X Laptop (128 GB), NVIDIA RTX Spark N1X Laptop (32 GB), NVIDIA Jetson Orin Nano 8GB (Super).
12 GB
12 GB tier →fits · 4.5 GB free · up to 32K ctx
16 GPUs · 1 device family
RTX 3080 Ti~337Simulate output on RTX 3080 TiRTX 3080 12GB~337Simulate output on RTX 3080 12GBRTX 5070~309Simulate output on RTX 5070iPhone~61Simulate output on Apple iPhone 17 Pro20 GB
fits · 12.5 GB free · up to 32K ctx
1 GPU
RX 7900 XT~112Simulate output on RX 7900 XT32–40 GB
fits · 24.5 GB free · up to 32K ctx
1 GPU · 3 Mac classes · 9 device families
RTX 5090~386Simulate output on RTX 5090M5 Max*36 GB~106Simulate output on Mac Studio M5 Max (36 GB)M632 GB~85Simulate output on iMac 24" M6 (32 GB)RTX Spark N1X~221Simulate output on NVIDIA RTX Spark N1X Laptop (32 GB)Orin~178Simulate output on NVIDIA Jetson AGX Orin 32GB
By VRAM tier · typical GPUs and Macs at Q4_K_M
Similar models
6| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| Qwen3.5 9BAlibaba | 99.4 | 6.6 GB | ~103 | same size |
| Gemma 2 9B ITGoogle | 68.0 | 7.3 GB | ~107 | same size |
| Ornith 1.5 9Bornith-ai | — | 6.6 GB | ~103 | same size |
| MiniCPM O 4 5OpenBMB | — | 8.3 GB | ~105 | same size |
| GLM 4.1V 9B ThinkingZ.ai | — | 7.2 GB | ~100 | same size |
| Shieldgemma 9BGoogle | — | 8.8 GB | ~107 | same size |
Specifications
- Publisher
- Zyphra
- Parameters
- 9.7B
- Active parameters
- 0.7B per token
- Architecture
- Zaya1VLForConditionalGeneration
- Context Length
- 32,768 tokens
- Vocabulary Size
- 262,272
- Release Date
- 2026-04-20
- License
- Apache 2.0
Frequently Asked Questions
How much VRAM does ZAYA1 VL 8B need?
ZAYA1 VL 8B requires 6.3 GB of VRAM at Q4_K_M, or 19.9 GB at BF16. Full 33K context adds up to 2.5 GB (8.8 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 9.7B × 4.8 bits ÷ 8 = 5.8 GB
KV Cache + Overhead ≈ 0.5 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 ≈ 3 GB (at full 33K context)
VRAM usage by quantization
What's the best quantization for ZAYA1 VL 8B?
For ZAYA1 VL 8B, Q4_K_M (6.3 GB) offers the best balance of quality and VRAM usage. Q5_K_M (7.4 GB) provides better quality if you have the VRAM. The smallest option is Q2_K at 4.6 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run ZAYA1 VL 8B on a Mac?
Yes — Mac Mini M2 Pro (16 GB) and 145 other Macs can run ZAYA1 VL 8B. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 6.3 GB of usable unified memory (RAM minus macOS overhead).
Can I run ZAYA1 VL 8B locally?
Yes — ZAYA1 VL 8B can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is ZAYA1 VL 8B?
At Q4_K_M, ZAYA1 VL 8B can reach ~124 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~346 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 → 0.5 GB active ÷ (8000 × 0.65) = 0.09 ms, plus 40 layers × 0.055 ms = 2.20 ms, so 1000 ÷ 2.29 ms = ~437 tok/s
Estimated speed at Q4_K_M (6.3 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of ZAYA1 VL 8B?
At Q4_K_M, the download is about 5.83 GB. The full-precision BF16 version is 19.44 GB. The smallest option (Q2_K) is 4.13 GB.
Which GPUs can run ZAYA1 VL 8B?
95 consumer GPUs can run ZAYA1 VL 8B at Q4_K_M. Top options include AMD Radeon RX 6700 XT, AMD Radeon RX 6700M, AMD Radeon RX 6750 XT, AMD Radeon RX 5700 XT. 51 GPUs have plenty of headroom for comfortable inference.
Which devices can run ZAYA1 VL 8B?
172 devices with unified memory can run ZAYA1 VL 8B at Q4_K_M (6.3 GB), including AMD Ryzen 7 8845HS (Hawk Point) Laptop, AMD Ryzen AI 7 350 (Krackan Point) Laptop, AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10. Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.