Hy3-preview

Checkpointtencent/Hy3-previewModel card indexedModel card

Support summary

Hy3-preview is a checkpoint published by Tencent, built on the HYV3ForCausalLM architecture. vLLM 0.28.0, the newest indexed vLLM release, documents support. Official recipes or cookbooks cover SGLang, without pinning a stable release.

vLLM documents support in 14 of 20 indexed releases, from 0.20.0 to 0.28.0.

Documented hardware
NVIDIA / CUDA and AMD / ROCm
Documented quantization
BF16
Documented features
Multi-token prediction, Reasoning parsers, Speculative decoding, Tensor parallelism, and Tool calling
Configured context
262,144 tokens
License
other

Hardware, quantization and features count only statements about this checkpoint; engine-wide capabilities are listed with each release.

Hugging Face model card
Publisher
Tencent
Task
text generation
License
other
Library
transformers
Architecture
HYV3ForCausalLM
Weight tensor elements
298.8B
Tensor types
BF16, F32
Configured context
262,144 tokens
Weight access
Public
Repository created
2026-04-13
Source 1: tencent/Hy3-preview · publisher model cardSource 2: tencent/Hy3-preview · Hugging Face metadataSource 3: tencent/Hy3-preview · publisher configurationReviewed
Model notes and sources

Engine support

3 engines

Engine support for Hy3-preview
EngineReleaseCheckpoint supportEvidence scopeLast known supported
vLLM0.28.0Newest indexedCheckpoint evidenceSource: vLLM v0.28.0 · docs/models/supported_models.mdReviewed 0.28.0This release
Features, dependencies and evidence for vLLM 0.28.0
SGLangVersion not pinnedRecipe / cookbookDocumented recipe

Official recipe

Source 1: sglang · docs/cookbook/autoregressive/Tencent/Hunyuan3-Preview.mdxSource 2: sglang · docs/src/snippets/autoregressive/hunyuan3-preview-deployment.jsxReviewed
No supported release indexed
Features, dependencies and evidence for SGLang 0.5.19
TensorRT-LLM1.2.1Newest indexedNo matching evidence curatedNo supported release indexed
Features, dependencies and evidence for TensorRT-LLM 1.2.1

Recipes and cookbooks