AI BENCHY 失敗分析
不正解 の失敗
どのAIモデルで 不正解 が起きやすいかを確認し、選ぶ前に信頼性のリスクを見極められます。
| 順位 | モデル | 企業 | 不正解 件数 | スコア | 正解テスト | 応答時間（平均） |
|---|---|---|---|---|---|---|
| #22 | Hy3 preview high | Tencent | 3 | 8.0 | 16/20 | 56.8s |
| #25 | Qwen3.5-27B medium | Qwen | 3 | 7.9 | 13/20 | 60.8s |
| #40 | MiMo-V2.5-Pro medium | Xiaomi | 3 | 7.6 | 12/20 | 21.8s |
| #46 | Claude Sonnet 4.6 medium | Anthropic | 3 | 7.6 | 13/20 | 15.7s |
| #48 | Gemini 3.1 Flash Lite high | 3 | 7.5 | 11/18 | 62.0s | |
| #54 | Kimi K2.6 medium | Moonshot AI | 3 | 7.4 | 12/20 | 54.1s |
| #57 | GPT-5.2 medium | OpenAI | 3 | 7.4 | 12/20 | 16.4s |
| #63 | Claude Opus 4.6 medium | Anthropic | 3 | 7.2 | 12/20 | 25.4s |
| #70 | Qwen3.5-35B-A3B medium | Qwen | 3 | 7.0 | 10/20 | 69.4s |
| #151 | Qwen3.5-9B medium | Qwen | 3 | 4.2 | 3/20 | 80.1s |
| #3 | Gemini 3.5 Flash low | 2 | 9.3 | 18/20 | 2.98s | |
| #4 | Gemini 3.1 Pro Preview medium | 2 | 9.3 | 18/20 | 20.8s | |
| #6 | Gemini 3.5 Flash medium | 2 | 9.0 | 17/20 | 4.29s | |
| #7 | Claude Opus 4.7 medium | Anthropic | 2 | 8.9 | 17/20 | 4.50s |
| #23 | Gemma 4 31B medium | 2 | 8.0 | 14/20 | 35.5s |