Tencent open-sources Hy4 Flash for low-cost multimodal workloads
Tencent released and open-sourced the Hy4 Flash family, comprising Hy4 Flash, Hy4 Flash-FP8 and Hy4 Flash Base, with downloadable weights for local deployment. The models natively accept text and images, support context windows of up to 1M tokens and activate 12B parameters per token; Tencent’s materials describe total size as either 224B or 220B. In PingWest’s WorkBuddy comparison, Hy4 Flash used the fewest calls and tokens for a presentation workflow, while the complete tested workflow was estimated to cost RMB 1.66, although it was not the most token-efficient model in a separate mobile-web task. Why it matters: Hy4 Flash combines downloadable weights, native multimodality, a 1M-token context window and lower API pricing with deployment across multiple Tencent products, making it directly available for developers building cost-sensitive office and agent workflows. What do you think?
Open the headline: source, AI briefing and more →Comments
…