Tencent Hunyuan released downloadable weights for three variants of Hy4 flash on November 18, two days after the model's API-only debut, giving developers deployable checkpoints across GitHub, Hugging Face Inc., ModelScope, and GitCode1,2.
The three variants are Hy4 flash, the instruction-tuned model; Hy4 flash-FP8, its quantized instruction-tuned counterpart; and Hy4-flash-base, a pretrained base checkpoint intended for further training rather than direct chat use.
Weight sizes and deployment stack
The standard checkpoint's weight-index file records approximately 440.5 GB of tensor data, while the FP8 variant records approximately 287.1 GB. These are weight-data sizes, not minimum GPU-memory requirements; quantization reduces storage but does not turn the release into a small consumer-device model.
Tencent provides dedicated vLLM and SGLang container images, with deployment examples using eight-way tensor parallelism. The vLLM recipe calls for host RAM because Engram tables are offloaded to CPU memory by default. The starting configuration uses an 8,192-token context, and the model's advertised 1M-token capacity is not a guarantee that every hardware configuration can serve that length. Tencent's Transformers instructions point to a model-support pull request rather than assuming an already-released package supports the architecture.
Pandaily checked all three official Hugging Face repositories and confirmed that weight-index files referenced available shard listings and that anonymous byte-range requests returned weight-file data for each variant; the outlet noted these checks establish download availability but are not full-checkpoint downloads or successful inference tests.
Licensing and what comes next
The checkpoints use Tencent Hy License 1.0, which Tencent says is not Apache 2.0 despite being derived from it. Section 10 of the license requires a separate written license for model-as-a-service operators exceeding $50 million in aggregate worldwide revenue over a consecutive 12-month period, or using the model in a product or service with more than 100 million monthly active users. A narrowly defined internal-use exception applies when the model and its outputs are not provided to third parties.
Tencent Hunyuan said it is accelerating work to open-source the larger Hy4 model, whose weight release is still being prepared. Earlier plans for Hy4 included visual understanding capabilities.
ANALYSIS The 440.5 GB standard checkpoint and eight-GPU tensor-parallelism requirement place Hy4 flash squarely in multi-node or high-end server territory, limiting self-hosted deployment to teams with substantial infrastructure. The revenue and MAU thresholds in the license add a commercial constraint that distinguishes this release from permissive open-weight competitors.