As the frontier of foundation models pushes toward ever-larger parameter counts and ever-longer context windows, the question is no longer just how much we can scale, but how efficiently we can do so.
Hy4-flash
tencent·Instruct·released 17 Nov 2026
hf download tencent/Hy4-flashHy4 flash is our most cost-effective Hy model to date, featuring 220B total parameters (including 64B Engram parameters) with just 12B active parameters per token. It natively accepts text and image inputs, generates text, and supports a context window of up to 1M tokens.
How much VRAM does Hy4-flash need?
At Q4_K_M Hy4-flash needs about 134 GB with an 8K context (134 GB at Q4, 434 GB at full precision). Pick your GPU to see what fits and how fast it runs.
Needs: ≈168 GB · Q4_K_M · 8K context — Pick a GPU to check the fit.
Advanced: context 8K · KV cache 16-bit
Smallest GPU for each quant and context (KV cache 16-bit)
| Quantisation | 8K | 16K | 32K | 64K | 128K | 1M |
|---|---|---|---|---|---|---|
| Q3_K_M | ||||||
| Q4_K_M | ||||||
| Q5_K_M | ||||||
| Q6_K | ||||||
| Q8_0 | ||||||
| FP16 |
≤16 GB card24–32 GB48–96 GBmore
| Quantisation | Total |
|---|---|
| Q2_KHeavy lossSmallest, noticeably worse quality | ≈74 GB |
| Q3_K_MNoticeable lossSmall, some quality loss | ≈108 GB |
| Q4_K_MrecommendedSmall lossRecommended default for local use | ≈168 GB |
| Q5_K_MNear-losslessNear-lossless for most tasks | ≈156 GB |
| Q6_KNear-losslessPractically lossless | ≈181 GB |
| Q8_0LosslessLossless in practice, twice the size of Q4 | ≈232 GB |
| FP16 / BF16LosslessFull precision, as released | ≈434 GB |
Estimates for 8K tokens of context: weights + KV cache + runtime overhead. All experts stay in memory, but only the active ones are read per token. Speed assumes memory-bound decoding at your GPU’s bandwidth. Pick a GPU above and the fit column checks every row against it. Weights marked “file” use the real GGUF download size; “est.” rows are computed from the parameter count.
How to pick a quant
Take the biggest quant that fits with room left for context. Q4_K_M is the usual sweet spot. Below Q4 quality drops fast, above Q6 you mostly pay memory for nothing. Big models (70B+, large MoE) tolerate low quants better than small ones.
- LosslessF16, Q8_0
- Near-losslessQ6_K, Q5_K_M
- Small lossQ5_K_S, Q4_K_M, IQ4_NL, IQ4_XS, Q4_K_S
- Noticeable lossQ4_0, Q3_K_L, Q3_K_M, IQ3_M
- Heavy lossIQ3_XXS, Q3_K_S, IQ2_M, Q2_K
Typical quality loss, not measured per model. Tiers come from KL divergence against FP16 on Llama 3 8B: llama.cpp perplexity README: LLaMA 3 8B scoreboard (KLD and ΔPPL vs FP16, Wikitext-2); llama.cpp quantize tool: ΔPPL per quant type @ Llama-3-8B
How much context fits?
At 8K context, Q4_K_M needs ≈168 GB(157 GB weights + 2.6 GB KV cache + 8.3 GB overhead). Pick a GPU above to see how much context fits.
Weights + overheadKV cache
Can the Apple M3 Ultra (512 GB) run Hy4-flash? →
Speed vs quality per quant
Smaller quants run faster and fit in less memory, but lose quality.
Fits in VRAMTight fitCPU offload✕ won't runQuality is typical KL divergence vs FP16, not measured for this model.
Real-world speed of Hy4-flash
Tokens per second measured by people running it at home. A GPU shows a median once 3 reports agree on the same setup. Reports far from our estimate wait for a review. Measure yours with llama-bench →
No medians yet. Be one of the first 3 to report a setup.
All reports
Run Hy4-flash locally
These commands load the original weights from Hugging Face.
from transformers import pipeline
pipe = pipeline("image-text-to-text", model="tencent/Hy4-flash", dtype="auto", device_map="auto")
messages = [
{
"role": "user",
"content": [
{"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bee.jpg"},
{"type": "text", "text": "Describe this image."},
],
}
]
out = pipe(text=messages, max_new_tokens=256)
print(out[0]["generated_text"][-1]["content"])Quantised versions and downloads
| Repository | Format | Quant | Size |
|---|---|---|---|
| Avifenesh/Hy4-flash-NVFP4 | Safetensors | NVFP4 | — |
| AngelSlim/Hy4-flash-GGUF | GGUF | IQ2_AH | 135 GB |
| AngelSlim/Hy4-flash-GGUF | GGUF | Q3_K | 155 GB |
| AngelSlim/Hy4-flash-GGUF | GGUF | Q4_K_M | 168 GB |
| Xylolabs/Hy4-flash-GGUF | GGUF | — | — |
| nace-ai/Hy4-flash-GGUF | GGUF | — | — |
| AtomicChat/Hy4-flash-GGUF | GGUF | — | — |
| TensorFold/Hy4-flash-MLX-4bit-MTP | MLX | 4BIT | — |
| TensorFold/Hy4-flash-MLX-8bit-MTP | MLX | 8BIT | — |
| apetersson/Hy4-flash-oQh8-BF16-Engram | MLX | BF16 | — |
About Hy4-flash
Hy4 flash is our most cost-effective Hy model to date, featuring 220B total parameters (including 64B Engram parameters) with just 12B active parameters per token. It natively accepts text and image inputs, generates text, and supports a context window of up to 1M tokens. Its 42-layer language backbone consists of one dense feed-forward layer and 41 MoE layers, with identity Hyper-Connections (iHC) supporting inter-layer information flow. A vision encoder (adapted from SigLIP2-So400m-Patch16-Res512) is connected to the backbone through an MLP projector. In addition to the backbone, a built-in 5-layer drafter, DFly2 (0.85B parameters), reuses the target model's KV cache and sparse indices to support long-context speculative decoding at a low, constant cost. Compared with Hy4, Hy4 Flash introduces three notable architectural changes: • Hybrid attention. Sparse attention (IndexPool) and sliding-window attention (SWA) are interleaved at a 1:3 ratio. • LatentMoE. Routed experts operate in a lower-dimensional latent space. Each token activates 16 of 512 routed experts, alongside one shared expert. • N-gram embedding (Engram). The backbone is augmented with learned bigram and trigram representations, retrieved via multi-head hash lookups. Bits-per-byte (BPB) on internal code & agent validation, lower is better: Few-shot pretrain benchmark evaluation:
Full model card on Hugging Face →Alternatives to Hy4-flash
All alternatives →MiniMax-M3.1-Flash is a native multimodal mixture-of-experts model for coding, agentic workflows, computer use, and cybersecurity. It supports text, image, and video inputs with a context window of up to 1M tokens.
👋 Join our WeChat or Discord community. 📖 Check out the GLM-5.3-Flash blog and GLM-5 Technical report . 📍 Use GLM-5.3-Flash API services on Z.ai API Platform.
Inkling-Small is a general-purpose multimodal model that accepts text, image and audio inputs and generates text outputs. It is intended for use in English and other languages, and across multiple coding languages.
Meet Qwen3-VL — the most powerful vision-language model in the Qwen series to date.
Command A+ is an open source model with 25 billion active parameters and 218B total parameters model optimized for agentic, multilingual, and reasoning-heavy tasks with a focus on enterprise performance, while also providing support for vision inputs for processing image inputs.
Hy4-flash FAQ
- How much VRAM does Hy4-flash need?
- About 134 GB at Q4_K_M with an 8K context. Q8_0 needs about 232 GB and full precision about 434 GB. Longer contexts need more memory for the KV cache.
- Can Hy4-flash run on the Apple M3 Ultra (512 GB)?
- Yes. It fits at Q8_0 (about 232 GB), at an estimated 1.9 tokens/s.
- Can I use Hy4-flash commercially?
- Unclear. It is released under tencent-hy-license. Custom licence: read the model card before using it commercially. Check the licence on the model card before using it in a product.
- What is the context length of Hy4-flash?
- 1,048,576 tokens (1M). The full window needs far more memory than an 8K context.
- How do I run Hy4-flash locally?
- It runs with Transformers. Each has copy-paste examples in its docs.
Rate this model
Data from Hugging Face, the model card and public leaderboards. Memory and speed figures are estimates; real numbers depend on the runtime, context length and drivers. Always check the licence yourself before commercial use.
0 comments