Introducing Hy4 Preview

Tencent released Hy4 Preview, a large language model with a significant size increase from its predecessor Hy3, which had 295B total parameters, 21B active parameters, 256,000 context length, and 598GB size. The new model supports two reasoning effort levels: 'high' (default) and 'no_think' (reasoning disabled), as shown in its chat template on Hugging Face. Simon Willison tested the model via OpenRouter with a prompt to generate an SVG of a pelican riding a bicycle, noting the reasoning trace uses truncated English for token efficiency.

This is a big size increase from their previous Hy3 https://huggingface.co/tencent/Hy3 in July, which was 295B, 21B active, 256,000 context, 598GB. I recently started using model chat templates to better understand their capabilities. Here's Hy4's chat template.jinja https://huggingface.co/tencent/Hy4-preview/blob/main/chat template.jinja on Hugging Face, which includes this section: {%- if not reasoning effort is defined %} {%- set reasoning effort = 'high' %} {%- elif reasoning effort not in 'high', 'no think' %} {%- if reasoning effort is none %} {{- raise exception 'reasoning effort error : None, should be no think/high' }} {%- else %} {{- raise exception 'reasoning effort error : ' + reasoning effort + ', should be no think/high' }} {%- endif %} {%- endif %} So it looks like there are just two reasoning effort levels: "high" the default and "no think" reason by disabled . I tried my "Generate an SVG of a pelican riding a bicycle" prompt with the default high reasoning via OpenRouter https://openrouter.ai/tencent/hy4-preview apps and got this https://tools.simonwillison.net/markdown-svg-renderer url=https%3A%2F%2Fgist.github.com%2Fsimonw%2Fcb69816b3fb940f2782569a82a523af1 : Quoting the reasoning trace: ... Let's maybe add a helmet? It could improve riding theme, but may obscure head. Maybe a small cycling cap or helmet? The user didn't ask; can add red helmet? Might be cute. But pelican with big beak; a helmet might obscure. Better maybe no. Maybe add sunglasses? no. Maybe add water? no. It's interesting how the reasoning trace uses slightly truncated English, presumably because perfect grammar isn't useful or token efficient for hidden reasoning text. Tags: ai https://simonwillison.net/tags/ai , generative-ai https://simonwillison.net/tags/generative-ai , llms https://simonwillison.net/tags/llms , pelican-riding-a-bicycle https://simonwillison.net/tags/pelican-riding-a-bicycle , llm-reasoning https://simonwillison.net/tags/llm-reasoning , llm-release https://simonwillison.net/tags/llm-release , ai-in-china https://simonwillison.net/tags/ai-in-china