Hy4 (Tencent)

Tencent released Hy4 on November 4, 2026, an open-weight model with roughly 770 billion parameters that adds visual understanding and computer operation to earlier models in the line. The release emphasizes three capability groups: fine-grained image processing, complex chart analysis and mathematical visual reasoning; computer use via both command-line and graphical interfaces; and training on professional software, with Blender and CAD tools named. The widely repeated claim that Hy4 ranks second among Chinese AI models is a Tencent marketing claim rather than an established fact, and the article separates vendor-reported results from verified ones. Tencent's announcement does support a 9-of-12 scenario lead on its WorkBuddy evaluation and a finance score of 3.04, though these benchmark figures have not been independently reproduced.

Tencent's Hy4 Preview is an open-weight text model with 770 billion total parameters, 49 billion active in a mixture-of-experts design, and a one-million-token context window, a large step up from the July Hy3 release at 295 billion parameters and 256K context. The weights occupy about 1.56 TB on Hugging Face. Inspecting the chat_template.jinja file, Simon Willison found the model expects a reasoning effort setting of either high, which is the default, or no think, indicating a built-in chain-of-thought layer. The model is blind rather than multimodal, and the post notes the hidden reasoning trace sometimes produces truncated, ungrammatical English.

Tencent released Hy4 preview, an open-weight MoE model from the Tencent Hy Team on August 28, 2026, with 770 billion total parameters of which 49 billion activate per token. The model supports roughly a one-million-token context window and includes a built-in multi-token prediction layer intended for speculative decoding. Weights ship under the Apache License 2.0 in both full-precision and FP8-quantized checkpoints distributed via Hugging Face, ModelScope, GitCode, and CNB. The analysis cautions that Tencent's comparative benchmarks came from an internally conducted blind evaluation, so the model's standing should be read as a self-selected snapshot rather than third-party-audited ranking.

Artificial Analysis lists Tencent Hy4 as a November 2026 model with reasoning, text and image input, text output, and a 1M-token context window. Its Intelligence Index v5.0.1 score is 35, above the comparable-model median of 30, although the page labels Hy4 proprietary and says no public API is available. The profile records 6.08B input tokens and 160M output tokens during the Intelligence Index evaluation, describing Hy4 as highly verbose against a 75M-token median. The benchmark applies to the displayed reasoning version, while the page only suggests that a non-reasoning variant may exist.