Leaked AI Benchmark Report Photo

"type": "photograph of a computer monitor displaying an academic technical report", "style": "slightly angled screen photo, visible moire pattern, LCD pixel grid, slight glare, LaTeX document formatting, serif fonts", "document_header": "left": "4 Benchmark Evaluation", "right": "argument name=\"report title\" default=\"DeepSeek-V4 Technical Report\"" , "introductory_text": "Paragraph summarizing comprehensive evaluation of argument name=\"main model name\" default=\"DeepSeek-V4\" against argument name=\"competitor model 1\" default=\"GPT-5.3\", argument name=\"competitor model 2\" default=\"Claude Opus 4.6\", and argument name=\"competitor model 3\" default=\"Gemini 3.1 Pro Preview\".", "visualizations": "legend": "5 items with color codes: dark blue, grey, light grey, blue striped, light blue", "bar_charts": "count": 6, "labels": [ "MMLU-Pro (EM)", "GPQA-Diamond (Pass@1)", "AIME 2025 (Pass@1)", "LiveCodeBench (Pass@1-COT)", "SWE-bench Verified (Resolved)", "Tau-bench (Average)" ] , "caption": "Figure 1 | Performance comparison on core benchmarks. DeepSeek-V4 achieves state-of-the-art results across the majority of benchmarks." , "data_table": "columns": [ "Benchmark", "argument name=\"main model name\" default=\"DeepSeek-V4\"", "argument name=\"competitor model 1\" default=\"GPT-5.3\"", "argument name=\"competitor model 2\" default=\"Claude Opus 4.6\"", "argument name=\"competitor model 3\" default=\"Gemini 3.1 Pro Preview\"", "GPT-4.1" ], "categories": "count": 4, "rows": [ "label": "General", "icon": "globe/network", "sub_items": 3, "label": "Reasoning & Math", "icon": "calculator/clipboard", "sub_items": 3, "label": "Code", "icon": "code brackets", "sub_items": 3, "label": "Agent", "icon": "robot face", "sub_items": 3 ]