Tencent Hy4 Outperforms GLM-5.3 and Kimi K3 in Blind Testing, with Output Pricing 82% Lower

Tencent Hy4 Outperforms GLM-5.3 and Kimi K3 in Blind Testing, with Output Pricing 82% Lower

2026-08-28 14:50

ChainThink News, August 28: According to Tencent's official announcement, its Hy4 preview model scored higher than GLM-5.3 and Kimi K3 in internal engineering blind tests.

163 Tencent internal experts tested the model using 203 real-world engineering tasks, with Hy4 achieving an average score of 2.99/4, surpassing GLM-5.3's 2.92/4 and Kimi K3's 2.94/4.

In direct comparisons, Hy4 achieved a win rate of 46.8% against GLM-5.3 (draws: 12.8%, losses: 40.4%), and a win rate of 51.2% against Kimi K3 (draws: 7.9%, losses: 40.9%).

However, Hy4 did not lead comprehensively in public benchmark scores; it still trails behind GLM-5.3 in code and cybersecurity evaluations such as DeepSWE and CyberGym.

Price remains Hy4’s primary advantage: input cost at ¥6 per million tokens and output at ¥18 per million tokens, representing 25% and approximately 36% savings compared to GLM-5.3, and 70% and 82% savings compared to Kimi K3, respectively.

Cache hit cost is only ¥0.3, 85% cheaper than GLM-5.3 and Kimi K3's ¥2.

Tencent Hy4 Outperforms GLM-5.3 and Kimi K3 in Blind Testing, with Output Pricing 82% Lower

Disclaimer: Contains third-party opinions, does not constitute financial advice

Recommended Reading

Sources: Bank of America exits underwriting talks for Moonshot (Kimi) IPO

3 hours ago
Sources: Bank of America exits underwriting talks for Moonshot (Kimi) IPO

Anthropic to reveal Q3 revenue today, paving the way for $2 trillion valuation IPO

6 hours ago
Anthropic to reveal Q3 revenue today, paving the way for $2 trillion valuation IPO

Meta Unveils Muse Wonder Series of AI Models, Shares Surge Over 3%

1 day ago
Meta Unveils Muse Wonder Series of AI Models, Shares Surge Over 3%

Anthropic Enhances Distillation Attack Defense, Has Suspended Tens of Thousands of Related Accounts

1 day ago
Anthropic Enhances Distillation Attack Defense, Has Suspended Tens of Thousands of Related Accounts

Pre-market News Summary: TSMC's Quarterly Earnings Poised to Hit Record High, Norway Imposes Nearly $10 Million Fine on Microsoft, Amazon in Talks to Acquire Decart

1 day ago
Pre-market News Summary: TSMC's Quarterly Earnings Poised to Hit Record High, Norway Imposes Nearly $10 Million Fine on Microsoft, Amazon in Talks to Acquire Decart

DeepSeek to Lead Pre-Series A Round for Winverse AI, Founded by Former Tencent AI Scientist, Post-Money Valuation of Approximately $400 Million

1 day ago
DeepSeek to Lead Pre-Series A Round for Winverse AI, Founded by Former Tencent AI Scientist, Post-Money Valuation of Approximately $400 Million

A whale longed for Lobster, initially floating a profit of $2.19 million, ultimately exited with a loss of $907,000.

4 days ago
A whale longed for Lobster, initially floating a profit of $2.19 million, ultimately exited with a loss of $907,000.