Tencent also published a real engineering blind test for Hy4 preview. 163 Tencent engineers used 203 engineering tasks to evaluate different models. Hy4 ended up with an average score of 2.99/4, above GLM-5.3's 2.92/4 and Kimi K3's 2.94/4.
Specifically, Hy4 beat GLM-5.3 46.8% of the time, tied 12.8%, and lost 40.4%. Against Kimi K3, it won 51.2%, tied 7.9%, and lost 40.9%. The tasks came from Tencent's real internal engineering work, not a fixed public benchmark.
However, Hy4 isn't sweeping public benchmarks. In Tencent's own evaluations, it trades wins and losses with models like GLM-5.3 and Kimi K3, and still trails GLM-5.3 on code and cybersecurity tests like DeepSWE and CyberGym.
Price is Hy4's most obvious advantage. At 6 yuan per million input tokens and 18 yuan per million output tokens, it's 25% and about 36% cheaper than GLM-5.3, respectively—and 70% and 82% cheaper than Kimi K3. A cache hit costs just 0.3 yuan, 85% less than the 2 yuan charged by both GLM-5.3 and Kimi K3.
Dongcha Beating Telegram channel · Dongcha Beating discussion group