News on November 8: As a new wave of Chinese AI large models enters the final countdown to release, core benchmark scores for Alibaba's Qwen4 series have recently been revealed ahead of time. In a specialized creativity evaluation for game development, some models in the series successfully surpassed the top overseas model Opus5.5, showing a remarkable pace of technological catch-up.
Game Development Creativity Benchmark Scores Revealed: Flash Successfully Surpasses Opus5.5
The benchmark revealed this time is not a traditional general-purpose large model benchmark, but a creativity assessment project focused on game development called GameFeel Bench.
Test data shows that the lightweight Qwen4-Flash scored 49.9 points, successfully surpassing Opus5.5xHigh, which scored 47.2. This is also the first time a Chinese large model has surpassed an overseas rival of this caliber in a related benchmark. Meanwhile, the smaller, locally deployable version Qwen4-27B also scored 44.6, performing almost as well as Sonnet5.5, which scored 45.
Ultimate Cost-Effectiveness and a Rich Product Matrix
In addition to its performance breakthrough, the Qwen4 series also has a clear advantage in inference usage costs.
According to leaked information, the cost for Qwen4-Flash to complete an equivalent task is only about $0.4, far lower than the $2.6 required by Opus5.5; Qwen4-27B's execution cost is even reduced to $0.16, making it extremely cost-competitive among products in the same tier.
