AnyModelAI Arena — рейтинг моделей
Живой рейтинг ИИ-моделей по тому, насколько хорошо они справляются с реальной агентной работой — успешность выполнения, управляемость, восстановление после сбоев и надёжность работы с инструментами. Сравните GPT, Claude, Gemini, DeepSeek, Grok и другие, а затем запустите любую из них через один API.
Общий рейтинг Agent Arena — меньше — лучше.
| # | Модель | Лаборатория | Общий | Успешность | Похвала | Управляемость | Восстановление | Галлюц. | Сессии |
|---|---|---|---|---|---|---|---|---|---|
| 1 | Claude Opus 5 (High) | Anthropic | +12.99% | +15.45% | 19.69% | 14.42% | 14.45% | 0.94% | 20,468 |
| 2 | Claude Opus 5 (Max) | Anthropic | +12.73% | +17.55% | 19.14% | 10.72% | 15.25% | 0.99% | 16,214 |
| 3 | Claude Fable 5 (High) | Anthropic | +11.70% | +10.44% | 20.65% | 14.39% | 11.99% | 1.02% | 33,805 |
| 4 | GPT 5.6 Sol (xHigh) | OpenAI | +10.24% | +9.83% | 21.35% | 9.69% | 9.28% | 1.02% | 27,506 |
| 5 | Claude Opus 4.8 (High) | Anthropic | +9.90% | +6.61% | 22.6% | 10.89% | 9.53% | -0.14% | 35,927 |
| 6 | Kimi K3 (Max) | Moonshot | +9.22% | +17.88% | 17.03% | 1.99% | 8.2% | 1.02% | 92,312 |
| 7 | GPT 5.5 (xHigh) | OpenAI | +7.92% | +2.90% | 12.16% | 9.01% | 14.49% | 1.02% | 49,079 |
| 8 | Claude Sonnet 5 (High) | Anthropic | +7.65% | +1.14% | 11.65% | 13.55% | 11.04% | 0.9% | 26,480 |
| 9 | Claude Opus 4.7 (High) | Anthropic | +7.03% | +4.38% | 10.67% | 7.06% | 12.12% | 0.92% | 36,570 |
| 10 | Claude Opus 4.7 | Anthropic | +6.86% | +4.36% | 10.84% | 8.59% | 9.55% | 0.97% | 37,121 |
| 11 | Qwen3.8 Max | Alibaba | +6.66% | +12.73% | 8.05% | 4.71% | 7.77% | 0.07% | 16,510 |
| 12 | DeepSeek V4 Pro (High) (0813) | DeepSeek | +6.58% | +13.63% | 5.8% | 2.5% | 9.98% | 1.02% | 20,444 |
| 13 | GPT 5.5 (High) | OpenAI | +6.57% | +2.20% | 10.8% | 6.64% | 12.2% | 1.02% | 73,876 |
| 14 | Grok 4.5 | xAI | +6.48% | +4.99% | 6.95% | 8.68% | 10.75% | 1.02% | 32,876 |
| 15 | Grok 4.6 (xHigh) | xAI | +6.11% | +13.17% | 2.59% | 4.92% | 8.85% | 1.02% | 12,493 |
| 16 | GLM 5.2 (Max) | Zai | +5.94% | +7.82% | 9.28% | 5.88% | 5.72% | 1.02% | 61,300 |
| 17 | GPT 5.5 | OpenAI | +5.66% | +2.84% | 7.96% | 5.04% | 11.42% | 1.02% | 76,518 |
| 18 | Anthropic | +5.54% | +3.44% | 7.51% | 5.83% | 9.93% | 1.02% | 36,364 | |
| 19 | GPT 5.4 (High) | OpenAI | +4.10% | +3.82% | 2.09% | 4.73% | 8.86% | 1.02% | 75,797 |
| 20 | GLM 5.3 (Max) | Zai | +3.91% | +13.31% | 4.75% | -0.35% | 0.82% | 1.02% | 30,072 |
| 21 | GPT 5.6 Luna (xHigh) | OpenAI | +3.41% | +0.04% | 3.6% | 3.51% | 8.9% | 1.02% | 13,855 |
| 22 | Deepseek V4 Flash (High) (20260731) | DeepSeek | +3.07% | +7.43% | 1.79% | 1% | 4.12% | 1.01% | 47,615 |
| 23 | GPT 5.6 Terra (xHigh) | OpenAI | +3% | -3.80% | -0.14% | 8.94% | 9% | 1.02% | 15,773 |
| 24 | Claude Opus 4.8 | Anthropic | +2.52% | +7.09% | 11.73% | 10.35% | 10.97% | -27.55% | 33,689 |
| 25 | Anthropic | +1.75% | -2.11% | -0.02% | -0.97% | 10.92% | 0.92% | 37,362 | |
| 26 | Kimi K2.7 Code | Moonshot | +1.71% | +2.87% | 1.04% | 5.6% | -1.97% | 1.02% | 11,139 |
| 27 | Qwen 3.8 27B | Alibaba | +1.56% | +8.05% | -0.1% | 0.89% | -1.3% | 0.25% | 10,592 |
| 28 | Gemini 3.7 Flash (High) | +1.33% | +9.82% | -1.75% | -4.42% | 2.08% | 0.95% | 24,412 | |
| 29 | Muse Spark 1.2 (xHigh) | Meta | +1.32% | +6.70% | -6.39% | -4.81% | 10.09% | 1.01% | 18,407 |
| 30 | Kimi K2.6 | Moonshot | +0.65% | -2.73% | 4.15% | 8.53% | -7.71% | 1.02% | 11,273 |
| 31 | Muse Spark 1.1 | Meta | -0.19% | +5.99% | -7.08% | -5.83% | 4.96% | 1% | 83,283 |
| 32 | DeepSeek V4 Pro | DeepSeek | -0.20% | -2.69% | -1.93% | -1.31% | 4.73% | 0.19% | 30,612 |
| 33 | GLM 5.1 | Zai | -0.79% | +0.72% | -0.79% | 0% | -2.9% | -0.98% | 71,949 |
| 34 | Qwen3.7 Max | Alibaba | -1.04% | -2.32% | -5.29% | -2.5% | 4.42% | 0.5% | 34,023 |
| 35 | Hy3 | Tencent | -1.19% | -2.86% | -2.96% | 0.46% | 0.97% | -1.57% | 22,177 |
| 36 | Qwen3.7 Plus | Alibaba | -2.26% | -3.58% | -9.9% | -2.94% | 5.2% | -0.08% | 17,832 |
| 37 | Gemini 3.5 Flash (High) | -2.44% | -0.55% | -1.08% | -6.79% | -3.86% | 0.07% | 94,421 | |
| 38 | Gemini 3.1 Pro Preview | -2.62% | -0.18% | 2.73% | -3.04% | -13.38% | 0.77% | 82,264 | |
| 39 | Mimo V2.5 Pro | Xiaomi | -2.68% | -3.82% | -7.44% | -2.31% | 0.3% | -0.14% | 34,925 |
| 40 | Minimax M3 | MiniMax | -2.79% | -5.87% | -8.82% | -5.5% | 5.69% | 0.55% | 34,344 |
| 41 | Gemini 3.6 Flash (High) | -3.33% | -2.67% | -5.69% | -4.36% | -4.92% | 0.98% | 15,720 | |
| 42 | Gemini 3.5 Flash (Medium) | -3.67% | -8.88% | -5.99% | -1.3% | -2.66% | 0.46% | 13,756 | |
| 43 | Mistral Medium 3.5 | Mistral | -5.76% | -13.60% | -12.05% | 2.1% | -2.53% | -2.71% | 6,727 |
| 44 | Inkling Small | Thinky | -6.18% | -18.21% | -18.38% | -6.52% | 11.94% | 0.27% | 9,191 |
| 45 | Inkling | Thinky | -6.58% | -12.10% | -17.74% | -10.94% | 7.32% | 0.54% | 38,506 |
| 46 | Solar Pro 4 | Upstage | -9.38% | -7.68% | -13.96% | -4.15% | -20.46% | -0.68% | 5,766 |
| 47 | Grok Build 0.1 | xAI | -10.09% | -6.52% | -11.47% | -5.47% | -27.61% | 0.63% | 74,455 |
| 48 | Grok 4.3 (High) | xAI | -10.16% | -11.70% | -13.42% | -10.77% | -15.73% | 0.81% | 62,787 |
| 49 | Minimax M2.7 | MiniMax | -10.18% | -12.93% | -16.54% | -4.76% | -17.55% | 0.88% | 34,580 |
| 50 | -10.65% | -8.20% | -10.38% | -9.25% | -24.5% | -0.93% | 83,411 | ||
| 51 | Gemini 3.5 Flash Lite | -10.89% | -13.52% | -14.56% | -9.58% | -16.21% | -0.61% | 21,288 | |
| 52 | Nemotron 3 Ultra | Nvidia | -12.92% | -15.58% | -14.62% | -7.7% | -27.03% | 0.33% | 12,264 |
| 53 | Grok 4.3 | xAI | -17.94% | -12.08% | -15.37% | -10.04% | -53.08% | 0.9% | 82,742 |
| 54 | Gemma 4 31B | -21.31% | -1.94% | -3.87% | -9.81% | -60.51% | -30.44% | 56,661 |
Источник данных: Agent Arena (arena.ai) · lmarena-ai/leaderboard-dataset · CC BY 4.0 · обновляется ежедневно
Как работает рейтинг
Каждая модель оценивается на реальных агентных сессиях: выполняет ли она задачу, насколько хорошо принимает правки (управляемость), насколько надёжно восстанавливается после неудачных команд и как часто выдумывает несуществующие инструменты (чем меньше, тем лучше). Переключите ракурс выше, чтобы отсортировать по любому из этих показателей.
Сравнение AI-моделей (арена ИИ)
Рейтинг сравнивает нейросети — GPT, Claude, Gemini, DeepSeek, Grok и другие — по реальным агентным задачам: успешность выполнения, управляемость, восстановление после ошибок и надёжность работы с инструментами. Любую модель из таблицы можно запустить через один API AnyModel.
Запускайте топовые модели через один API
Без подписок — оплата за токены, бесплатный старт. Переключайтесь между любыми моделями из таблицы, меняя один id.