ひっかけ問題を同一条件で投げ、264モデルの偏差値を出しました。生回答・判定・単価まで全公開しています。
偏差値 = (素点 - 平均) / 標準偏差 × 10 + 50。平均が50、標準偏差が10になるよう換算した値です。母集団はこの表に載っている264モデルで、満点は20点です。
同じ20問を同じ条件で出題。素点・偏差値・出力単価(1Mトークンあたり・円)の順。単価は GET /v1/models/pricing のライブ値です。
| # | モデル | 素点/20 | 偏差値 | 出力 ¥/1M |
|---|---|---|---|---|
| 1 | ~anthropic/claude-fable-latest | 20 | 66.3 | ¥7,350 |
| 2 | ~anthropic/claude-opus-latest | 20 | 66.3 | ¥3,675 |
| 3 | anthropic/claude-fable-5 | 20 | 66.3 | ¥7,350 |
| 4 | anthropic/claude-fable-5.1 | 20 | 66.3 | ¥7,350 |
| 5 | anthropic/claude-opus-5 | 20 | 66.3 | ¥3,675 |
| 6 | anthropic/claude-opus-5-fast | 20 | 66.3 | ¥7,868 |
| 7 | openrouter/auto | 20 | 66.3 | ¥472 |
| 8 | shitate/orchestrator:max | 20 | 66.3 | ¥8,261 |
| 9 | o4-mini | 19 | 64.4 | ¥647 |
| 10 | claude-fable-5 | 19 | 64.4 | ¥7,350 |
| 11 | gemini-2.5-pro | 19 | 64.4 | ¥1,470 |
| 12 | deepseek-reasoner | 19 | 64.4 | ¥345 |
| 13 | moonshotai/kimi-k2.5 | 19 | 64.4 | ¥331 |
| 14 | openai/gpt-5.6-luna | 19 | 64.4 | ¥176 |
| 15 | openai/gpt-5.6-sol | 19 | 64.4 | ¥1,470 |
| 16 | google/gemini-3.1-pro-preview | 19 | 64.4 | ¥1,764 |
| 17 | deepseek/deepseek-v4-flash | 19 | 64.4 | ¥104 |
| 18 | qwen/qwen3.8-2.4t-a95b | 19 | 64.4 | ¥882 |
| 19 | openai/gpt-6-astra | 19 | 64.4 | ¥7,350 |
| 20 | google/gemini-3.8-flash | 19 | 64.4 | ¥551 |
| 21 | openai/gpt-5.3-codex | 19 | 64.4 | ¥2,058 |
| 22 | openai/gpt-5-mini | 19 | 64.4 | ¥294 |
| 23 | openai/o3 | 19 | 64.4 | ¥1,176 |
| 24 | ~google/gemini-pro-latest | 19 | 64.4 | ¥1,764 |
| 25 | teai/lux | 19 | 64.4 | ¥7,868 |
| 26 | ~openai/gpt-latest | 19 | 64.4 | – |
| 27 | deepseek/deepseek-v4-flash-vision-exp | 19 | 64.4 | ¥97 |
| 28 | google/gemini-3.1-pro-preview-customtools | 19 | 64.4 | ¥1,764 |
| 29 | google/gemini-3.7-flash | 19 | 64.4 | ¥295 |
| 30 | openai/gpt-5.5 | 19 | 64.4 | ¥4,410 |
| 31 | openai/gpt-5.6-sol-pro | 19 | 64.4 | ¥1,470 |
| 32 | openai/gpt-6-astra-pro | 19 | 64.4 | ¥7,350 |
| 33 | shitate/orchestrator | 19 | 64.4 | ¥209 |
| 34 | shitate/orchestrator:solid | 19 | 64.4 | ¥4,546 |
| 35 | claude-opus-4-8 | 18 | 62.6 | ¥3,675 |
| 36 | google/gemini-3-flash-preview | 18 | 62.6 | ¥94 |
| 37 | deepseek/deepseek-v4-pro | 18 | 62.6 | ¥312 |
| 38 | deepseek/deepseek-v4-flash-0731 | 18 | 62.6 | ¥16 |
| 39 | moonshotai/kimi-k3 | 18 | 62.6 | ¥1,953 |
| 40 | nvidia/nemotron-3-ultra-550b-a55b | 18 | 62.6 | ¥353 |
| 41 | sakana/sakana-namazu | 18 | 62.6 | ¥588 |
| 42 | z-ai/glm-5.1 | 18 | 62.6 | ¥446 |
| 43 | bytedance-seed/seed-2.0-code | 18 | 62.6 | ¥441 |
| 44 | anthropic/claude-opus-4.7 | 18 | 62.6 | ¥3,675 |
| 45 | openai/gpt-5.1-codex | 18 | 62.6 | ¥1,470 |
| 46 | openai/o4-mini | 18 | 62.6 | ¥647 |
| 47 | teai/auto | 18 | 62.6 | ¥2,360 |
| 48 | ~google/gemini-flash-latest | 18 | 62.6 | ¥551 |
| 49 | teai/fast | 18 | 62.6 | ¥104 |
| 50 | openai/gpt-5.6-terra-pro | 18 | 62.6 | ¥1,180 |
| 51 | shitate/orchestrator-max | 18 | 62.6 | ¥4,736 |
| 52 | teai/max | 18 | 62.6 | ¥4,736 |
| 53 | claude-opus-4-7 | 17 | 60.7 | ¥3,675 |
| 54 | claude-sonnet-4-5-20250929 | 17 | 60.7 | ¥2,360 |
| 55 | google/gemini-3.5-flash | 17 | 60.7 | ¥708 |
| 56 | google/gemma-4-31b-it | 17 | 60.7 | ¥50 |
| 57 | anthropic/claude-opus-4.8 | 17 | 60.7 | ¥3,675 |
| 58 | openai/o1 | 17 | 60.7 | ¥8,820 |
| 59 | ~moonshotai/kimi-latest | 17 | 60.7 | ¥1,610 |
| 60 | ~openai/gpt-mini-latest | 17 | 60.7 | ¥662 |
20問の生回答JSON(全264モデル) — 1問ごとの回答と正誤が入っています。
外科医の罠・川渡りの罠・最長単語の罠の3問。モデル数は多いぶん母集団が違うので、上の表とは数字を比較しないでください。網羅性を見るための表です。
| # | モデル | 素点/3 | 偏差値 | 出力 ¥/1M |
|---|---|---|---|---|
| 1 | claude-fable-5 | 3 | 68.8 | ¥7,350 |
| 2 | claude-sonnet-4-5-20250929 | 3 | 68.8 | ¥2,360 |
| 3 | qwen/qwen3-235b-a22b | 3 | 68.8 | ¥268 |
| 4 | mistralai/mistral-small-3.2-24b-instruct | 3 | 68.8 | ¥29 |
| 5 | microsoft/phi-4 | 3 | 68.8 | ¥21 |
| 6 | openai/gpt-5.6-luna-pro | 3 | 68.8 | ¥176 |
| 7 | nvidia/Nemotron-3-Nano-30B-A3B | 3 | 68.8 | ¥25 |
| 8 | anthropic/claude-opus-5 | 3 | 68.8 | ¥3,675 |
| 9 | z-ai/glm-5.3-flash | 3 | 68.8 | ¥39 |
| 10 | qwen/qwen3.8-flash | 3 | 68.8 | ¥69 |
| 11 | tencent/hy-mt2-30b-a3b | 3 | 68.8 | ¥43 |
| 12 | tencent/hy-mt2-7b | 3 | 68.8 | ¥43 |
| 13 | anthropic/claude-fable-5.1 | 3 | 68.8 | ¥7,350 |
| 14 | aion-labs/aion-3.0-mini | 3 | 68.8 | ¥206 |
| 15 | x-ai/grok-build-0.1 | 3 | 68.8 | ¥294 |
| 16 | aion-labs/aion-2.0 | 3 | 68.8 | ¥235 |
| 17 | mistralai/ministral-3b-2512 | 3 | 68.8 | ¥15 |
| 18 | openai/gpt-5.1-codex-mini | 3 | 68.8 | ¥294 |
| 19 | qwen/qwen3-next-80b-a3b-thinking | 3 | 68.8 | ¥176 |
| 20 | z-ai/glm-4.5-air | 3 | 68.8 | ¥125 |
| 21 | anthropic/claude-sonnet-4 | 3 | 68.8 | ¥2,205 |
| 22 | thedrummer/skyfall-36b-v2 | 3 | 68.8 | ¥118 |
| 23 | amazon/nova-lite-v1 | 3 | 68.8 | ¥35 |
| 24 | openai/gpt-4o-2024-05-13 | 3 | 68.8 | ¥2,205 |
| 25 | teai/auto | 3 | 68.8 | ¥2,360 |
| 26 | anthropic/claude-opus-5-fast | 3 | 68.8 | ¥7,868 |
| 27 | teai/lux | 3 | 68.8 | ¥7,868 |
| 28 | gpt-4o-mini | 2 | 57.8 | ¥88 |
| 29 | gpt-4.1 | 2 | 57.8 | ¥1,176 |
| 30 | gpt-4o | 2 | 57.8 | ¥1,470 |
| 31 | gpt-4.1-mini | 2 | 57.8 | ¥235 |
| 32 | claude-opus-4-7 | 2 | 57.8 | ¥3,675 |
| 33 | gemini-2.5-flash-lite | 2 | 57.8 | ¥59 |
| 34 | claude-haiku-4-5-20251001 | 2 | 57.8 | ¥787 |
| 35 | claude-sonnet-4-6 | 2 | 57.8 | ¥2,205 |
| 36 | claude-opus-4-5 | 2 | 57.8 | ¥3,675 |
| 37 | claude-opus-4-6 | 2 | 57.8 | ¥3,675 |
| 38 | gemini-2.5-pro | 2 | 57.8 | ¥1,470 |
| 39 | meta-llama/llama-4-maverick | 2 | 57.8 | ¥96 |
| 40 | google/gemini-3-flash-preview | 2 | 57.8 | ¥94 |
| 41 | qwen/qwen3.5-flash-02-23 | 2 | 57.8 | ¥38 |
| 42 | qwen/qwen3-coder | 2 | 57.8 | ¥147 |
| 43 | mistralai/codestral-2508 | 2 | 57.8 | ¥132 |
| 44 | cohere/command-a | 2 | 57.8 | ¥1,470 |
| 45 | qwen/qwen3.5-plus-02-15 | 2 | 57.8 | ¥229 |
| 46 | openai/gpt-5.6-luna | 2 | 57.8 | ¥176 |
| 47 | openai/gpt-5.4 | 2 | 57.8 | ¥2,205 |
| 48 | openai/gpt-5.4-mini | 2 | 57.8 | ¥662 |
| 49 | openai/gpt-5.6-sol | 2 | 57.8 | ¥1,470 |
| 50 | openai/gpt-5.4-nano | 2 | 57.8 | ¥184 |
| 51 | openai/gpt-5.5 | 2 | 57.8 | ¥4,410 |
| 52 | google/gemini-3.1-flash-lite | 2 | 57.8 | ¥118 |
| 53 | openai/gpt-5.6-sol-pro | 2 | 57.8 | ¥1,470 |
| 54 | openai/gpt-5.6-terra-pro | 2 | 57.8 | ¥1,180 |
| 55 | google/gemini-3.5-flash | 2 | 57.8 | ¥708 |
| 56 | google/gemini-3.1-pro-preview | 2 | 57.8 | ¥1,764 |
| 57 | z-ai/glm-5.2 | 2 | 57.8 | ¥152 |
| 58 | x-ai/grok-4.3 | 2 | 57.8 | ¥368 |
| 59 | moonshotai/kimi-k3 | 2 | 57.8 | ¥1,953 |
| 60 | qwen/qwen3.8-27b | 2 | 57.8 | ¥375 |
temperature=0・同一の実行環境で投げています。逐次実行ではありませんが、同一条件を崩さない範囲で並列化しています。note に明記しています。temperature=0 でも回答がブレます。再取得したものは rerun に日付が入ります。