GPT-5 and Gemini were completely defeated before Oracle, and Tencent released its first ancient text evaluation benchmark, Chronicles OCR According to Beating monitoring, Tencent Hybrid and SSV Digital Culture Laboratory, together with institutions such as the Institute of Information Technology of the Chinese Academy of Sciences, have officially launched the first ancient text perception evaluation benchmark, Chronicles OCR, covering the "Seven Body Change". This benchmark includes 2800 images cross annotated by experts, and for the first time quantifies the recognition difficulty of seven fonts from oracle bone script to cursive script. The research team evaluated 28 mainstream multimodal big language models, and the results showed that they were almost completely wiped out on ancient fonts. In the cross era character detection task, the core metrics of GPT-5 and Gemini 2.5 Pro are close to 0, and the strongest performing model is only 16.5. Even if the positioning step is waived by directly framing the image, the highest accuracy is only 27.1%, with Gemini 3.1 Pro having an accuracy of only 14.0% on Oracle. This confirms that modern models heavily rely on regular modern layout priors. Faced with unconstrained and strongly noisy ancient physical media, the text segmentation mechanism of the model directly fails. The font classification results further indicate that the model often recognizes carrier textures (such as turtle shell or bronze rust) rather than real character strokes. The experiment also revealed a counterintuitive phenomenon: activating the thinking mode actually leads to a decrease in the recognition rate of ancient Chinese characters. Comparison shows that almost all models that support this mode exhibit degradation after enabling thinking. When the underlying visual perception is lacking, the thought chain not only fails to correct errors, but also becomes an illusion amplifier, outputting highly confident incorrect answers. [Original link]