🤖 Tencent releases budget model Hy4 Flash

Tencent unveiled the budget Hy4 flash on November 16: a 224-billion-parameter MoE (12 billion active) and 64 billion for Engram conditional memory. Context is 1 million tokens, speed exceeds 150 tokens/s. API prices per 1 million tokens: $0.070 input, $0.281 output, compared to $0.834 and $2.501 for Hy4.

🌍 In a blind test with 85 WorkBuddy users, satisfaction is comparable to Claude Haiku 5.5, with average task cost 20 times lower than Hy4 and 8 times lower than DeepSeek-V4.1-Flash-1017 (vendor measurements). This will increase pressure on DeepSeek, Alibaba, and Anthropic, while the promised open weights will provide a 224B model for self-hosting.

👤 The model is free in WorkBuddy until November 30, with a 50% API discount. It is suitable for office tasks, search, slides, and simple frontend demos; for complex code and long agent chains, Tencent recommends Hy4.

Source 1: https://hy.tencent.com/research/hy4-flash Source 2: https://www.chaincatcher.com/en/article/2305090