LLM Benchmark Evaluation Source-mixed reference results; not a controlled head-to-head comparison Agnes-3.0-Flash Qwen3.6-35B-A3B Kimi K2.5 Muse Glimmer Qwen3.5-27B DeepSeek V4 Flash Qwen3.8-27B Gemini 3.5 Flash Qwen3.8 Flash Next MiniMax M3 85.05 84.1 78.9 83.5 85.8 90.8 90.5 92.2 92.3 92.9 GPQA Diamond 68.33 66.7 59 80 72.3 79.7 82 81 79.7 74 AA-LCR 74.2 64.4 43.7 77 75.6 75.8 79.5 76.3 81.3 82.9 IFBench 38.08 35.8 39.6 43.6 39.5 50.3 46.6 53.1 50.6 45.4 SciCode 23 18.8 22.9 27 20.7 40.4 18.4 51.4 24.5 16.7 AA-Omniscience Accuracy