KingBench 3
A comprehensive evaluation benchmark for large language models covering multiple dimensions including simulation, 3D interaction, SVG rendering, mathematical reasoning, and fine-tuning.
Timeline (last 90 days)
In the UP host's KingBench 3 eight-item test, GPT-6 Sol scored 66/80 (82.5%) and Opus 5.5 scored 75/80 (93.75%), with Opus leading by 11.25 percentage points.
Qwen3 Max 0902 scored a perfect 10 points on the 3D watch problem in KingBench 3, the first perfect score any model has achieved on that problem.
Compared to the previous version, Qwen3 Max 0902 lost points on three problems — elevator simulation, contact lens case, and archery shooting — and gained points only on the 3D watch problem.
On the 3D watch problem, the previous all-time high score was 8 points, set by Muse Spark 1.3.
Fable 5.1 scored 74/80 (92.5%) in the KingBench 3 small-scale test, which is 2 points higher than Astra.
In the 8 KingBench 3 small-scale tests, Astra won on the 3D folding table, panda eating a burger SVG, and 3D wristwatch tests, while Fable had the advantage on the elevator, glasses case, and archery tests.
GPT-6 Astra scored a high 90% (72/80) on the KingBench 3 benchmark.
All Facts (5)
In the UP host's KingBench 3 eight-item test, GPT-6 Sol scored 66/80 (82.5%) and Opus 5.5 scored 75/80 (93.75%), with Opus leading by 11.25 percentage points.
50%UnverifiedGPT-6 Astra scored a high 90% (72/80) on the KingBench 3 benchmark.
50%UnverifiedFable 5.1 scored 74/80 (92.5%) in the KingBench 3 small-scale test, which is 2 points higher than Astra.
50%UnverifiedIn the 8 KingBench 3 small-scale tests, Astra won on the 3D folding table, panda eating a burger SVG, and 3D wristwatch tests, while Fable had the advantage on the elevator, glasses case, and archery tests.
50%Unverified在博主的KingBench基准测试中，Qwen3该版本获得65/80分（81.25%），位列第二名
50%