Tencent releases Hy4 flash model weights for local deployment
Read more
TechNode
technode.com

Tencent releases Hy4 flash model weights for local deployment

Tencent announced the weights of its artificial intelligence model, Hy4 flash, on Wednesday. This release came two days after the model was introduced to users via the company's applications and cloud API. Developers can now download the standard version of the model, its FP8 format version, as well as the base model for further training.

This release aligns with Tencent's previous statement regarding plans to make the model weights publicly available. The Hy4 flash model is capable of processing both text and image input, and its context window reaches one million tokens.

The model has approximately 224 billion parameters, with about 12 billion parameters activated per token. The main goal of Hy4 flash is to improve daily performance in tasks such as document preparation, research, and spreadsheet work. Tencent accompanied the weights with relevant deployment materials, distributing them under the Tencent Hy license.

Similar stories

Tencent releases Hy4 model weights with a million-token context window
Read more
technode.com

Tencent releases Hy4 model weights with a million-token context window

Tencent announced on Friday the weights of its flagship artificial intelligence model, Hy4, allowing developers to download and implement the model independently. This release follows the initial launch of Hy4 on November 4th and makes the flagship model available under the Apache 2.0 license, which permits both commercial use and modification.

The Hy4 model is capable of understanding text, images, and video, featuring a context window of approximately one million tokens. Its 'Mixture of Experts' language framework contains 770 billion parameters, with 49 billion parameters activated for processing each token.

According to Tencent's statement, this version of the model surpasses the Hy4 preview in areas such as instruction following, long-context understanding, advanced task execution, and scientific research.

The model supports workflows covering graphical user interfaces and command-line tools, as well as frontend development, game design, and office tasks. Tencent provides deployment instructions for vLLM and SGLang, along with resources for fine-tuning and a safety evaluation map. The release of the flagship model followed the separate unveiling of the smaller Hy4 flash model.

Tencent releases Hy4 flash model weights for download in three variants
Read more
pandaily.com

Tencent releases Hy4 flash model weights for download in three variants

Tencent has published the downloadable weights for the Hy4 flash model following the announcement of its API on November 16th. This was reported by IT Home on November 18th. This update provides deployable checkpoints, rather than a re-presentation of the API specifications and pricing previously covered by Pandaily.

The official Tencent model page lists three versions: Hy4 flash, which is an instruction-tuned model; Hy4 flash-FP8, its instruction-tuned quantized version; and Hy4 flash-base, a pre-trained base model. The base checkpoint is intended for further training, not to replace the chat-ready variants.

Pandaily verified all three official repositories on Hugging Face. The weight index files referenced available shard lists, and anonymous byte range requests returned weight file data for each version. These checks confirm the availability for download but do not constitute full checkpoint downloads or successful output tests.

The standard checkpoint contains approximately 440.5 GB of tensor data, while the FP8 version is about 287.1 GB. These sizes refer to the weight data, not the minimum GPU memory requirements. Quantization reduces storage size but does not turn the release into a model suitable for small consumer devices.

Deployment documentation is an important part of this release. Tencent provides specialized vLLM and SGLang container images, including examples of eight-way tensor parallelism. Instructions for Transformers point to requesting model support rather than assuming that the standard released package already supports this architecture.

The vLLM recipe also requires host RAM, as Engram tables are defaulted to be offloaded to CPU memory. The initial configuration uses a context size of 8192 tokens; the model's claimed capacity of 1 million tokens does not guarantee that every hardware configuration can handle such length. The published memory hints are estimates, not measured minimum requirements.

The checkpoints use the Tencent Hy 1.0 license, which explicitly states that it is not Apache 2.0, despite being derived from it. Section 10 requires a separate written license for Model-as-a-Service operators whose cumulative global revenue exceeds $50 million over a consecutive 12-month period, or for using the model in a product or service with more than 100 million monthly active users. A narrowly defined exception applies to internal use when the model and its outputs are not provided to third parties.

Thus, this release is best described as open in terms of weights according to Tencent's stated terms, rather than as unrestricted open source. Tencent reports that the release of the larger Hy4 model weights is still in preparation.

Tencent introduces Hy4 flash model, reducing AI task costs by 95% in internal tests
Read more
technode.com

Tencent introduces Hy4 flash model, reducing AI task costs by 95% in internal tests

Tencent released Hy4 flash on Monday—a more cost-effective artificial intelligence model designed for everyday tasks such as creating presentations, conducting research, and editing spreadsheets.

According to Tencent's internal benchmark, WorkIndex, the cost of executing a single task with the Hy4 flash model was only one twentieth compared to the full Hy4 model. The API prices are set as follows: $0.07 per million input tokens, $0.281 per million output tokens, and $0.003 per million cached input tokens.

The Hy4 flash model supports multimodal input and has a context window of up to one million tokens. It possesses 224 billion parameters and activates 12 billion parameters during direct pass.

In a separate blind test, where 85 WorkBuddy users evaluated results across 259 tasks, Tencent reported that the model achieved a user satisfaction level close to Claude Haiku 5.5, while having the lowest cost and shortest completion time among the tested models.

The model is available in WorkBuddy and Yuanbao, as well as via TokenHub on Tencent Cloud. WorkBuddy users can use it for free until November 30th, and TokenHub offers API access at half the standard price. Tencent plans to publish the model weights and make it available on OpenRouter.

Popular