Tencent Hy4 Preview Gets Optimization Upgrade: Same Quality, Fewer Turns, Lower Tokens
Key Info
Tencent's Hy4 preview has shipped an optimization update targeting long thinking and over-verification on complex tasks. The model now delivers the same task quality with fewer turns and lower input/output token usage, and the update is live for everyone.
Highlights
- Optimization focuses on reducing excessive reasoning and verification loops on complex tasks.
- Same quality with lower cost: fewer interaction turns and reduced input/output tokens, confirmed by benchmarks and human evaluations.
- Available now through WorkBuddy; further fast iterations are planned. The underlying model remains a 770B-parameter model with 49B active parameters and 1M context.