Hy4 preview
Points and comments are a snapshot, not live.
Tencent open-sources Hy4 preview, a 770B-parameter LLM for productivity tasks.
Tencent released Hy4 preview, a 770B total parameter (49B active) large language model with a 1M+ token context window. It is open-source and available via WorkBuddy, CodeBuddy, Yuanbao, Tencent Cloud TokenHub, and OpenRouter. The model achieved an average score of 2.99/4.00 in a blind internal evaluation of 203 engineering tasks, slightly ahead of GLM-5.3 and Kimi K3. Hy4 preview contributed to its own training optimization via a self-improvement loop, and autonomously improved its inference throughput by 31.8%. API pricing is $0.834/M input tokens, $2.501/M output, and $0.042/M for cache hits.
What commenters are saying
Commenters noted Hy4's high token volume on OpenRouter, but some suspected Tencent may be gaming usage stats by buying its own tokens. Others praised Hy3's surprising quality for its price and noted a big jump between Hy3's preview and final releases. Several called for better cache cost metrics and sorting on OpenRouter. One commenter observed that Hy4's low cache price (5%) makes it more compelling than competitors with 10-20% cache costs.