tryaii-dre 0.2.0

package/dist/registry/presets/defaultModels.json

@@ -0,0 +1,435 @@
+{
+  "version": "0.1.0",
+  "updated": "2026-01",
+  "models": [
+    {
+      "model_id": "gpt-4o",
+      "provider": "OpenAI",
+      "benchmark_scores": {
+        "MMLU": 86.4, "HellaSwag": 95.3, "HumanEval": 90.2, "SWE-bench": 30.7,
+        "TruthfulQA": 71.5, "ARC": 85.0, "GSM8K": 92.0, "DROP": 83.4,
+        "SuperGLUE": 90.0, "Chatbot Arena (LMSys)": 1441, "MT-Bench": 9.0
+      },
+      "capabilities": ["vision", "web-search", "multimodal"],
+      "pricing": {"input_per_1k": 0.0025, "output_per_1k": 0.01},
+      "latency": "fast",
+      "description": "GPT-4o: High-capability multimodal model"
+    },
+    {
+      "model_id": "gpt-4o-mini",
+      "provider": "OpenAI",
+      "benchmark_scores": {
+        "MMLU": 70.0, "HellaSwag": 85.5, "HumanEval": 87.2, "SWE-bench": 8.7,
+        "TruthfulQA": 47.0, "ARC": 82.0, "GSM8K": 87.0, "DROP": 65.0,
+        "SuperGLUE": 70.0, "Chatbot Arena (LMSys)": 1315, "MT-Bench": 7.5
+      },
+      "capabilities": ["fast-response", "cost-effective"],
+      "pricing": {"input_per_1k": 0.00015, "output_per_1k": 0.0006},
+      "latency": "very fast",
+      "description": "GPT-4o Mini: Fast and cost-effective"
+    },
+    {
+      "model_id": "o1",
+      "provider": "OpenAI",
+      "benchmark_scores": {
+        "MMLU": 83.5, "HellaSwag": 95.0, "HumanEval": 92.0, "SWE-bench": 48.9,
+        "TruthfulQA": 50.0, "ARC": 85.0, "GSM8K": 95.0, "DROP": 60.0,
+        "SuperGLUE": 75.0, "Chatbot Arena (LMSys)": 1398, "MT-Bench": 8.0
+      },
+      "capabilities": ["advanced-reasoning", "complex-problem-solving"],
+      "pricing": {"input_per_1k": 0.015, "output_per_1k": 0.06},
+      "latency": "slow",
+      "description": "O1: Advanced reasoning model"
+    },
+    {
+      "model_id": "o3",
+      "provider": "OpenAI",
+      "benchmark_scores": {
+        "MMLU": 85.6, "HellaSwag": 94.0, "HumanEval": 88.0, "SWE-bench": 69.1,
+        "TruthfulQA": 60.0, "ARC": 92.0, "GSM8K": 90.0, "DROP": 80.0,
+        "SuperGLUE": 85.0, "Chatbot Arena (LMSys)": 1450, "MT-Bench": 9.0
+      },
+      "capabilities": ["advanced-reasoning", "latest-model"],
+      "pricing": {"input_per_1k": 0.01, "output_per_1k": 0.04},
+      "latency": "medium",
+      "description": "O3: Latest reasoning model"
+    },
+    {
+      "model_id": "o4-mini",
+      "provider": "OpenAI",
+      "benchmark_scores": {
+        "MMLU": 80.6, "HellaSwag": 90.0, "HumanEval": 70.0, "SWE-bench": 68.1,
+        "TruthfulQA": 55.0, "ARC": 85.0, "GSM8K": 80.0, "DROP": 75.0,
+        "SuperGLUE": 78.0, "Chatbot Arena (LMSys)": 1397, "MT-Bench": 8.5
+      },
+      "capabilities": ["advanced-reasoning", "fast-response"],
+      "pricing": {"input_per_1k": 0.00110, "output_per_1k": 0.0044},
+      "latency": "fast",
+      "description": "O4 Mini: Fast reasoning"
+    },
+    {
+      "model_id": "gpt-5",
+      "provider": "OpenAI",
+      "benchmark_scores": {
+        "MMLU": 91.0, "HellaSwag": 95.0, "HumanEval": 92.0, "SWE-bench": 74.9,
+        "TruthfulQA": 78.0, "ARC": 93.0, "GSM8K": 95.0, "DROP": 86.0,
+        "SuperGLUE": 90.0, "Chatbot Arena (LMSys)": 1461, "MT-Bench": 9.6
+      },
+      "capabilities": ["advanced-reasoning", "multimodal", "latest-model"],
+      "pricing": {"input_per_1k": 0.00125, "output_per_1k": 0.01},
+      "latency": "medium",
+      "description": "GPT-5: Flagship model"
+    },
+    {
+      "model_id": "gpt-5-mini",
+      "provider": "OpenAI",
+      "benchmark_scores": {
+        "MMLU": 82.0, "HellaSwag": 88.0, "HumanEval": 88.0, "SWE-bench": 71.0,
+        "TruthfulQA": 60.0, "ARC": 88.0, "GSM8K": 92.0, "DROP": 82.0,
+        "SuperGLUE": 78.0, "Chatbot Arena (LMSys)": 1392, "MT-Bench": 9.0
+      },
+      "capabilities": ["fast-response", "cost-effective"],
+      "pricing": {"input_per_1k": 0.000375, "output_per_1k": 0.0015},
+      "latency": "fast",
+      "description": "GPT-5 Mini: Efficient lightweight"
+    },
+    {
+      "model_id": "gpt-5.1",
+      "provider": "OpenAI",
+      "benchmark_scores": {
+        "MMLU": 90.8, "HellaSwag": 96.0, "HumanEval": 93.0, "SWE-bench": 76.5,
+        "TruthfulQA": 80.0, "ARC": 94.0, "GSM8K": 96.5, "DROP": 87.0,
+        "SuperGLUE": 91.0, "Chatbot Arena (LMSys)": 1475, "MT-Bench": 9.7, "LiveBench": 75.0
+      },
+      "capabilities": ["advanced-reasoning", "multimodal", "vision", "latest-model"],
+      "pricing": {"input_per_1k": 0.00125, "output_per_1k": 0.01},
+      "latency": "medium",
+      "description": "GPT-5.1: Enhanced flagship"
+    },
+    {
+      "model_id": "gpt-5.2",
+      "provider": "OpenAI",
+      "benchmark_scores": {
+        "MMLU": 92.0, "HellaSwag": 97.0, "HumanEval": 95.0, "SWE-bench": 79.5,
+        "TruthfulQA": 82.0, "ARC": 90.5, "GSM8K": 98.0, "DROP": 89.0,
+        "SuperGLUE": 93.0, "Chatbot Arena (LMSys)": 1490, "MT-Bench": 9.8, "LiveBench": 78.0
+      },
+      "capabilities": ["advanced-reasoning", "multimodal", "vision", "agentic", "latest-model"],
+      "pricing": {"input_per_1k": 0.00125, "output_per_1k": 0.01},
+      "latency": "medium",
+      "description": "GPT-5.2: Latest flagship with agentic capabilities"
+    },
+    {
+      "model_id": "gpt-4.1",
+      "provider": "OpenAI",
+      "benchmark_scores": {
+        "MMLU": 90.2, "HellaSwag": 98.0, "HumanEval": 90.0, "SWE-bench": 54.6,
+        "TruthfulQA": 75.0, "ARC": 88.0, "GSM8K": 97.0, "DROP": 85.4,
+        "SuperGLUE": 90.0, "Chatbot Arena (LMSys)": 1406, "MT-Bench": 9.2
+      },
+      "capabilities": ["advanced-reasoning", "multimodal"],
+      "pricing": {"input_per_1k": 0.002, "output_per_1k": 0.008},
+      "latency": "fast",
+      "description": "GPT-4.1: Refined general-purpose model"
+    },
+    {
+      "model_id": "gpt-4.1-nano",
+      "provider": "OpenAI",
+      "benchmark_scores": {
+        "MMLU": 80.1, "HellaSwag": 88.0, "HumanEval": 85.0, "SWE-bench": 20.0,
+        "TruthfulQA": 50.0, "ARC": 72.0, "GSM8K": 80.0, "DROP": 70.0,
+        "SuperGLUE": 70.0, "Chatbot Arena (LMSys)": 1320, "MT-Bench": 7.0
+      },
+      "capabilities": ["fast-response", "cost-effective"],
+      "pricing": {"input_per_1k": 0.0001, "output_per_1k": 0.0004},
+      "latency": "very fast",
+      "description": "GPT-4.1 Nano: Ultra-cheap, ultra-fast"
+    },
+    {
+      "model_id": "gpt-5-nano",
+      "provider": "OpenAI",
+      "benchmark_scores": {
+        "MMLU": 80.0, "HellaSwag": 85.0, "HumanEval": 80.0, "SWE-bench": 54.7,
+        "TruthfulQA": 50.0, "ARC": 75.0, "GSM8K": 90.0, "DROP": 75.0,
+        "SuperGLUE": 70.0, "Chatbot Arena (LMSys)": 1346, "MT-Bench": 8.0
+      },
+      "capabilities": ["fast-response", "cost-effective"],
+      "pricing": {"input_per_1k": 0.000075, "output_per_1k": 0.0003},
+      "latency": "very fast",
+      "description": "GPT-5 Nano: Smallest, cheapest GPT-5 variant"
+    },
+    {
+      "model_id": "claude-3-7-sonnet-20250219",
+      "provider": "Anthropic",
+      "benchmark_scores": {
+        "MMLU": 80.7, "HellaSwag": 80.0, "HumanEval": 85.0, "SWE-bench": 62.3,
+        "TruthfulQA": 65.0, "ARC": 84.0, "GSM8K": 90.0, "DROP": 80.0,
+        "SuperGLUE": 80.0, "Chatbot Arena (LMSys)": 1385, "MT-Bench": 8.5
+      },
+      "capabilities": ["long-context", "code-generation", "analysis"],
+      "pricing": {"input_per_1k": 0.003, "output_per_1k": 0.015},
+      "latency": "medium",
+      "description": "Claude 3.7 Sonnet: Strong all-rounder"
+    },
+    {
+      "model_id": "claude-sonnet-4-20250514",
+      "provider": "Anthropic",
+      "benchmark_scores": {
+        "MMLU": 86.5, "HellaSwag": 89.0, "HumanEval": 92.0, "SWE-bench": 72.7,
+        "TruthfulQA": 75.0, "ARC": 87.0, "GSM8K": 91.0, "DROP": 85.0,
+        "SuperGLUE": 90.0, "Chatbot Arena (LMSys)": 1400, "MT-Bench": 9.5
+      },
+      "capabilities": ["advanced-reasoning", "long-context", "code-generation"],
+      "pricing": {"input_per_1k": 0.003, "output_per_1k": 0.015},
+      "latency": "medium",
+      "description": "Claude Sonnet 4: Advanced coding and reasoning"
+    },
+    {
+      "model_id": "claude-sonnet-4-5-20250929",
+      "provider": "Anthropic",
+      "benchmark_scores": {
+        "MMLU": 89.1, "HellaSwag": 92.0, "HumanEval": 93.0, "SWE-bench": 77.2,
+        "TruthfulQA": 75.0, "ARC": 90.0, "GSM8K": 95.0, "DROP": 85.0,
+        "SuperGLUE": 90.0, "Chatbot Arena (LMSys)": 1446, "MT-Bench": 9.5, "LiveBench": 73.0
+      },
+      "capabilities": ["advanced-reasoning", "long-context"],
+      "pricing": {"input_per_1k": 0.003, "output_per_1k": 0.015},
+      "latency": "medium",
+      "description": "Claude Sonnet 4.5: Enhanced reasoning"
+    },
+    {
+      "model_id": "claude-haiku-4-5-20251001",
+      "provider": "Anthropic",
+      "benchmark_scores": {
+        "MMLU": 65.4, "HellaSwag": 82.0, "HumanEval": 88.3, "SWE-bench": 73.3,
+        "TruthfulQA": 55.0, "ARC": 82.0, "GSM8K": 85.0, "DROP": 85.0,
+        "SuperGLUE": 72.0, "Chatbot Arena (LMSys)": 1378, "MT-Bench": 7.5, "LiveBench": 71.38
+      },
+      "capabilities": ["fast-response", "cost-effective", "computer-use"],
+      "pricing": {"input_per_1k": 0.0008, "output_per_1k": 0.004},
+      "latency": "fast",
+      "description": "Claude Haiku 4.5: Fast and affordable"
+    },
+    {
+      "model_id": "claude-opus-4-5-20251101",
+      "provider": "Anthropic",
+      "benchmark_scores": {
+        "MMLU": 90.8, "HellaSwag": 97.0, "HumanEval": 94.0, "SWE-bench": 80.9,
+        "TruthfulQA": 78.0, "ARC": 93.0, "GSM8K": 96.0, "DROP": 88.0,
+        "SuperGLUE": 92.0, "Chatbot Arena (LMSys)": 1493, "MT-Bench": 9.7, "LiveBench": 76.0
+      },
+      "capabilities": ["advanced-reasoning", "long-context", "agentic", "computer-use", "code-generation"],
+      "pricing": {"input_per_1k": 0.015, "output_per_1k": 0.075},
+      "latency": "slow",
+      "description": "Claude Opus 4.5: Most capable Anthropic model"
+    },
+    {
+      "model_id": "gemini-2.5-pro",
+      "provider": "Google",
+      "benchmark_scores": {
+        "MMLU": 84.1, "HellaSwag": 95.0, "HumanEval": 90.0, "SWE-bench": 63.8,
+        "TruthfulQA": 72.0, "ARC": 91.0, "GSM8K": 96.0, "DROP": 85.0,
+        "SuperGLUE": 88.0, "Chatbot Arena (LMSys)": 1458, "MT-Bench": 9.3
+      },
+      "capabilities": ["multimodal", "long-context"],
+      "pricing": {"input_per_1k": 0.00075, "output_per_1k": 0.003},
+      "latency": "medium",
+      "description": "Gemini 2.5 Pro: Strong multimodal model"
+    },
+    {
+      "model_id": "gemini-2.0-flash",
+      "provider": "Google",
+      "benchmark_scores": {
+        "MMLU": 82.0, "HellaSwag": 90.0, "HumanEval": 80.0, "SWE-bench": 55.0,
+        "TruthfulQA": 68.0, "ARC": 83.0, "GSM8K": 90.0, "DROP": 80.0,
+        "SuperGLUE": 82.0, "Chatbot Arena (LMSys)": 1365, "MT-Bench": 8.5
+      },
+      "capabilities": ["fast-response", "multimodal"],
+      "pricing": {"input_per_1k": 0.0001, "output_per_1k": 0.0004},
+      "latency": "fast",
+      "description": "Gemini 2.0 Flash: Fast and affordable"
+    },
+    {
+      "model_id": "gemini-2.5-flash",
+      "provider": "Google",
+      "benchmark_scores": {
+        "MMLU": 84.0, "HellaSwag": 93.0, "HumanEval": 78.0, "SWE-bench": 50.0,
+        "TruthfulQA": 68.0, "ARC": 85.0, "GSM8K": 92.0, "DROP": 82.0,
+        "SuperGLUE": 85.0, "Chatbot Arena (LMSys)": 1406, "MT-Bench": 9.0
+      },
+      "capabilities": ["fast-response", "multimodal"],
+      "pricing": {"input_per_1k": 0.00015, "output_per_1k": 0.00035},
+      "latency": "fast",
+      "description": "Gemini 2.5 Flash: Balanced speed and quality"
+    },
+    {
+      "model_id": "gemini-2.5-flash-lite",
+      "provider": "Google",
+      "benchmark_scores": {
+        "MMLU": 84.5, "HellaSwag": 88.0, "HumanEval": 72.0, "SWE-bench": 45.0,
+        "TruthfulQA": 65.0, "ARC": 78.0, "GSM8K": 88.0, "DROP": 75.0,
+        "SuperGLUE": 80.0, "Chatbot Arena (LMSys)": 1340, "MT-Bench": 8.0
+      },
+      "capabilities": ["fast-response", "cost-effective", "multimodal"],
+      "pricing": {"input_per_1k": 0.000075, "output_per_1k": 0.0003},
+      "latency": "very fast",
+      "description": "Gemini 2.5 Flash Lite: Ultra-affordable"
+    },
+    {
+      "model_id": "gemini-3-pro-preview",
+      "provider": "Google",
+      "benchmark_scores": {
+        "MMLU": 91.8, "HellaSwag": 96.0, "HumanEval": 92.0, "SWE-bench": 76.2,
+        "TruthfulQA": 75.0, "ARC": 93.0, "GSM8K": 97.0, "DROP": 87.0,
+        "SuperGLUE": 91.0, "Chatbot Arena (LMSys)": 1501, "MT-Bench": 9.5, "LiveBench": 74.0
+      },
+      "capabilities": ["advanced-reasoning", "multimodal", "long-context", "agentic", "deep-think"],
+      "pricing": {"input_per_1k": 0.00075, "output_per_1k": 0.003},
+      "latency": "medium",
+      "description": "Gemini 3 Pro: Top-tier multimodal reasoning"
+    },
+    {
+      "model_id": "gemini-3-flash-preview",
+      "provider": "Google",
+      "benchmark_scores": {
+        "MMLU": 88.5, "HellaSwag": 94.0, "HumanEval": 88.0, "SWE-bench": 78.0,
+        "TruthfulQA": 72.0, "ARC": 88.0, "GSM8K": 94.0, "DROP": 84.0,
+        "SuperGLUE": 88.0, "Chatbot Arena (LMSys)": 1465, "MT-Bench": 9.2, "LiveBench": 72.0
+      },
+      "capabilities": ["fast-response", "multimodal", "agentic", "code-generation"],
+      "pricing": {"input_per_1k": 0.00015, "output_per_1k": 0.0006},
+      "latency": "fast",
+      "description": "Gemini 3 Flash: Fast agentic model"
+    },
+    {
+      "model_id": "deepseek-reasoner",
+      "provider": "DeepSeek",
+      "benchmark_scores": {
+        "MMLU": 90.8, "HellaSwag": 88.0, "HumanEval": 65.0, "SWE-bench": 49.2,
+        "TruthfulQA": 65.0, "ARC": 82.0, "GSM8K": 94.5, "DROP": 70.0,
+        "SuperGLUE": 80.0, "Chatbot Arena (LMSys)": 1394, "MT-Bench": 8.0
+      },
+      "capabilities": ["advanced-reasoning", "mathematical"],
+      "pricing": {"input_per_1k": 0.00055, "output_per_1k": 0.00219},
+      "latency": "medium",
+      "description": "DeepSeek Reasoner: Strong math and reasoning"
+    },
+    {
+      "model_id": "deepseek-chat",
+      "provider": "DeepSeek",
+      "benchmark_scores": {
+        "MMLU": 88.5, "HellaSwag": 88.9, "HumanEval": 65.2, "SWE-bench": 50.0,
+        "TruthfulQA": 60.0, "ARC": 78.0, "GSM8K": 89.3, "DROP": 80.0,
+        "SuperGLUE": 80.0, "Chatbot Arena (LMSys)": 1356, "MT-Bench": 8.5
+      },
+      "capabilities": ["cost-effective", "code-capable"],
+      "pricing": {"input_per_1k": 0.00027, "output_per_1k": 0.0011},
+      "latency": "fast",
+      "description": "DeepSeek Chat: Affordable general-purpose"
+    },
+    {
+      "model_id": "grok-3-latest",
+      "provider": "xAI",
+      "benchmark_scores": {
+        "MMLU": 92.7, "HellaSwag": 95.0, "HumanEval": 86.5, "SWE-bench": 60.0,
+        "TruthfulQA": 70.0, "ARC": 85.0, "GSM8K": 89.3, "DROP": 85.0,
+        "SuperGLUE": 88.0, "Chatbot Arena (LMSys)": 1409, "MT-Bench": 9.2
+      },
+      "capabilities": ["real-time-info", "web-access"],
+      "pricing": {"input_per_1k": 0.003, "output_per_1k": 0.015},
+      "latency": "medium",
+      "description": "Grok 3: Real-time information access"
+    },
+    {
+      "model_id": "grok-3-mini-latest",
+      "provider": "xAI",
+      "benchmark_scores": {
+        "MMLU": 78.9, "HellaSwag": 85.0, "HumanEval": 50.0, "SWE-bench": 40.0,
+        "TruthfulQA": 55.0, "ARC": 75.0, "GSM8K": 75.0, "DROP": 65.0,
+        "SuperGLUE": 70.0, "Chatbot Arena (LMSys)": 1362, "MT-Bench": 7.5
+      },
+      "capabilities": ["cost-effective", "real-time-info"],
+      "pricing": {"input_per_1k": 0.0003, "output_per_1k": 0.0005},
+      "latency": "fast",
+      "description": "Grok 3 Mini: Affordable with web access"
+    },
+    {
+      "model_id": "grok-4-latest",
+      "provider": "xAI",
+      "benchmark_scores": {
+        "MMLU": 92.1, "HellaSwag": 95.0, "HumanEval": 94.7, "SWE-bench": 75.0,
+        "TruthfulQA": 70.0, "ARC": 92.0, "GSM8K": 95.2, "DROP": 85.0,
+        "SuperGLUE": 88.0, "Chatbot Arena (LMSys)": 1480, "MT-Bench": 9.3
+      },
+      "capabilities": ["advanced-reasoning", "real-time-info", "web-access", "tool-use"],
+      "pricing": {"input_per_1k": 0.002, "output_per_1k": 0.01},
+      "latency": "medium",
+      "description": "Grok 4: Premium reasoning with web"
+    },
+    {
+      "model_id": "grok-4-fast",
+      "provider": "xAI",
+      "benchmark_scores": {
+        "MMLU": 90.5, "HellaSwag": 93.0, "HumanEval": 91.0, "SWE-bench": 70.0,
+        "TruthfulQA": 68.0, "ARC": 88.0, "GSM8K": 92.0, "DROP": 82.0,
+        "SuperGLUE": 85.0, "Chatbot Arena (LMSys)": 1450, "MT-Bench": 9.0
+      },
+      "capabilities": ["fast-response", "real-time-info", "web-access"],
+      "pricing": {"input_per_1k": 0.0005, "output_per_1k": 0.0025},
+      "latency": "fast",
+      "description": "Grok 4 Fast: Balanced speed and capability"
+    },
+    {
+      "model_id": "grok-4-1-fast-reasoning-latest",
+      "provider": "xAI",
+      "benchmark_scores": {
+        "MMLU": 94.2, "HellaSwag": 93.0, "HumanEval": 89.3, "SWE-bench": 68.0,
+        "TruthfulQA": 72.0, "ARC": 87.0, "GSM8K": 92.0, "DROP": 82.0,
+        "SuperGLUE": 86.0, "Chatbot Arena (LMSys)": 1455, "MT-Bench": 9.0, "LiveBench": 70.0
+      },
+      "capabilities": ["fast-response", "tool-use", "agentic", "real-time-info"],
+      "pricing": {"input_per_1k": 0.0005, "output_per_1k": 0.0025},
+      "latency": "fast",
+      "description": "Grok 4.1 Fast Reasoning: Fast agentic reasoning"
+    },
+    {
+      "model_id": "grok-code-fast",
+      "provider": "xAI",
+      "benchmark_scores": {
+        "MMLU": 82.0, "HellaSwag": 88.0, "HumanEval": 92.0, "SWE-bench": 70.8,
+        "TruthfulQA": 65.0, "ARC": 82.0, "GSM8K": 85.0, "DROP": 78.0,
+        "SuperGLUE": 80.0, "Chatbot Arena (LMSys)": 1400, "MT-Bench": 8.5, "LiveBench": 68.0
+      },
+      "capabilities": ["fast-response", "cost-effective", "agentic", "code-generation"],
+      "pricing": {"input_per_1k": 0.00015, "output_per_1k": 0.0006},
+      "latency": "fast",
+      "description": "Grok Code Fast: Specialized for coding tasks"
+    },
+    {
+      "model_id": "mistral-large-latest",
+      "provider": "Mistral",
+      "benchmark_scores": {
+        "MMLU": 84.0, "HellaSwag": 90.0, "HumanEval": 65.0, "SWE-bench": 40.0,
+        "TruthfulQA": 60.0, "ARC": 82.0, "GSM8K": 81.0, "DROP": 70.0,
+        "SuperGLUE": 75.0, "Chatbot Arena (LMSys)": 1313, "MT-Bench": 7.5
+      },
+      "capabilities": ["cost-effective", "code-capable"],
+      "pricing": {"input_per_1k": 0.002, "output_per_1k": 0.006},
+      "latency": "medium",
+      "description": "Mistral Large: European AI model"
+    },
+    {
+      "model_id": "mistral-small-latest",
+      "provider": "Mistral",
+      "benchmark_scores": {
+        "MMLU": 39.9, "HellaSwag": 70.0, "HumanEval": 30.0, "SWE-bench": 20.0,
+        "TruthfulQA": 40.0, "ARC": 72.0, "GSM8K": 66.7, "DROP": 50.0,
+        "SuperGLUE": 50.0, "Chatbot Arena (LMSys)": 1351, "MT-Bench": 6.0
+      },
+      "capabilities": ["cost-effective", "fast-response"],
+      "pricing": {"input_per_1k": 0.0001, "output_per_1k": 0.0003},
+      "latency": "very fast",
+      "description": "Mistral Small: Ultra-affordable"
+    }
+  ]
+}

package/dist/router.d.ts

@@ -0,0 +1,178 @@
+/**
+ * Main Router -- the primary public API for TryAii-DRE.
+ *
+ * Usage:
+ *   import { Router } from 'tryaii-dre';
+ *
+ *   const router = new Router();
+ *   const result = await router.route('Write a Python function to merge sorted arrays');
+ *   console.log(result.bestModel);   // e.g., "gpt-5.2"
+ *   console.log(result.scores);      // Top models with scores and reasoning
+ *
+ * The default `LocalEmbeddingProvider` is async-only, so `route()` itself is
+ * async. Callers that have injected a sync embedding provider can use the
+ * niche `routeSync()` method for a blocking call.
+ */
+import { BenchmarkRegistry } from './benchmarks/registry.js';
+import { ClassificationResult } from './classifiers/base.js';
+import { TryaiiDreConfig } from './config.js';
+import { BaseEmbeddingProvider } from './embeddings/base.js';
+import { ModelRegistry, ModelInfo } from './registry/models.js';
+import { ModelScore } from './scoring/engine.js';
+import { Priorities } from './scoring/priorities.js';
+import type { LatencyTier } from './types.js';
+/**
+ * Maximum allowed prompt length (characters). Prompts longer than this are
+ * truncated before classification to avoid OOM in the embedding model. Must
+ * stay in sync with the Python SDK's MAX_PROMPT_LENGTH (classifiers/base.py).
+ */
+export declare const MAX_PROMPT_LENGTH = 100000;
+/**
+ * Result of routing a prompt.
+ *
+ * Contains the recommended model, all scored models, and classification details.
+ */
+export interface RouteResult {
+    /** The top recommended model ID. */
+    bestModel: string;
+    /** All scored models (sorted by score descending). */
+    scores: ModelScore[];
+    /** Classification details (what kind of task was detected). */
+    classification: ClassificationResult | null;
+    /** Priorities used for this routing decision. */
+    priorities: Priorities;
+}
+/** Get list of model IDs in ranked order from a RouteResult. */
+export declare function routeResultTopK(result: RouteResult): string[];
+/** Get the score of the top model. */
+export declare function routeResultBestScore(result: RouteResult): number;
+/** Get the reasoning for why the top model was chosen. */
+export declare function routeResultBestReasoning(result: RouteResult): string;
+/** Options for route(). */
+export interface RouteOptions {
+    /** Quality/cost/speed priorities. Defaults to balanced. */
+    priorities?: Priorities;
+    /** Number of top models to return. */
+    topK?: number;
+    /** Only consider models from this provider. */
+    filterProvider?: string;
+    /** Only consider models with this capability. */
+    filterCapability?: string;
+    /** Only consider models cheaper than this (input $/1k tokens). */
+    filterMaxCost?: number;
+}
+/**
+ * Semantic AI model router.
+ *
+ * Analyzes user prompts using embeddings, matches them against benchmark
+ * centroids, and recommends the best AI model based on benchmark performance,
+ * pricing, latency, and user priorities.
+ */
+export declare class Router {
+    private _config;
+    private _registry;
+    private _benchmarkRegistry;
+    private _scoringEngine;
+    private _embeddingProvider;
+    private _centroidLoader;
+    private _classifier;
+    constructor(opts?: {
+        config?: Partial<TryaiiDreConfig>;
+        registry?: ModelRegistry;
+        benchmarkRegistry?: BenchmarkRegistry;
+        embeddingProvider?: BaseEmbeddingProvider;
+    });
+    /**
+     * Lazy-initialize the embedding provider + centroid loader.
+     *
+     * The loader is stored on `this` so that `addBenchmark()` mutates the
+     * same instance the classifier reads from -- custom benchmarks added
+     * via the Router become visible to subsequent `route()` calls without
+     * re-instantiating anything.
+     */
+    private _ensureCentroidLoader;
+    /** Lazy-initialize the embedding classifier on first use. */
+    private _ensureClassifier;
+    /**
+     * Route a prompt to the best AI model.
+     *
+     * Async by default -- works with any embedding provider, including the
+     * default async `LocalEmbeddingProvider`. For a blocking call backed by a
+     * sync provider, see `routeSync()`.
+     *
+     * @param prompt - The user's input text to classify and route.
+     * @param opts - Routing options (priorities, filters, topK).
+     * @returns RouteResult with the best model and full scoring breakdown.
+     */
+    route(prompt: string, opts?: RouteOptions): Promise<RouteResult>;
+    /**
+     * Synchronous version of `route()`.
+     *
+     * Requires the injected embedding provider to support sync calls
+     * (`supportsSync === true`). Throws otherwise -- the default
+     * `LocalEmbeddingProvider` is async-only, so calling `routeSync()` on a
+     * default `Router` will fail. Inject a sync provider (e.g. a custom
+     * cached provider) to use this path.
+     */
+    routeSync(prompt: string, opts?: RouteOptions): RouteResult;
+    /**
+     * Validate and normalize a prompt before classification.
+     *
+     * Rejects empty/non-string prompts and truncates to MAX_PROMPT_LENGTH,
+     * matching the Python Router (which raises ValueError on empty/non-string
+     * and slices to the same limit) so both SDKs behave identically.
+     */
+    private static _normalizePrompt;
+    /** Shared post-classification path: filter models, score, return RouteResult. */
+    private _buildResult;
+    /**
+     * Shortcut to add a model to the registry.
+     *
+     * @see ModelRegistry.add()
+     */
+    addModel(opts: {
+        modelId: string;
+        provider: string;
+        benchmarks?: Record<string, number>;
+        pricing?: [number, number];
+        latency?: LatencyTier;
+        capabilities?: string[];
+        description?: string;
+    }): ModelInfo;
+    /**
+     * Add a custom benchmark to the routing system.
+     *
+     * Async by default -- works with any embedding provider, including the
+     * async-only `LocalEmbeddingProvider`. The new centroid is generated
+     * immediately and added to the shared centroid loader, so subsequent
+     * `route()` calls see it without any restart.
+     *
+     * For sync-provider callers who want a blocking setup step, use
+     * `addBenchmarkSync()`.
+     *
+     * @param name - Benchmark name (e.g., "CustomerSupportQA").
+     * @param queries - Representative prompts for this benchmark (10-20 recommended).
+     * @param description - Human-readable description.
+     * @param minScore - Minimum score for normalization.
+     * @param maxScore - Maximum score for normalization.
+     */
+    addBenchmark(name: string, queries: string[], description?: string, minScore?: number, maxScore?: number): Promise<void>;
+    /**
+     * Synchronous version of `addBenchmark()`.
+     *
+     * Requires the injected embedding provider to support sync calls
+     * (`supportsSync === true`). The default `LocalEmbeddingProvider` is
+     * async-only, so calling `addBenchmarkSync()` on a default `Router`
+     * will throw.
+     */
+    addBenchmarkSync(name: string, queries: string[], description?: string, minScore?: number, maxScore?: number): void;
+    /** Register a benchmark in the registry and rebuild the scoring normalizer. */
+    private _registerBenchmark;
+    /** Access the model registry. */
+    get models(): ModelRegistry;
+    /** Access the benchmark registry. */
+    get benchmarks(): BenchmarkRegistry;
+    /** Access the configuration. */
+    get config(): TryaiiDreConfig;
+}
+//# sourceMappingURL=router.d.ts.map

package/dist/router.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"router.d.ts","sourceRoot":"","sources":["../src/router.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;GAcG;AAEH,OAAO,EAAE,iBAAiB,EAAuB,MAAM,0BAA0B,CAAC;AAGlF,OAAO,EAAE,oBAAoB,EAAE,MAAM,uBAAuB,CAAC;AAE7D,OAAO,EAAE,eAAe,EAAyC,MAAM,aAAa,CAAC;AACrF,OAAO,EAAE,qBAAqB,EAAE,MAAM,sBAAsB,CAAC;AAE7D,OAAO,EAAE,aAAa,EAAE,SAAS,EAAE,MAAM,sBAAsB,CAAC;AAChE,OAAO,EAAE,UAAU,EAAiB,MAAM,qBAAqB,CAAC;AAChE,OAAO,EAAsB,UAAU,EAAE,MAAM,yBAAyB,CAAC;AACzE,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,YAAY,CAAC;AAE9C;;;;GAIG;AACH,eAAO,MAAM,iBAAiB,SAAU,CAAC;AAEzC;;;;GAIG;AACH,MAAM,WAAW,WAAW;IAC1B,oCAAoC;IACpC,SAAS,EAAE,MAAM,CAAC;IAElB,sDAAsD;IACtD,MAAM,EAAE,UAAU,EAAE,CAAC;IAErB,+DAA+D;IAC/D,cAAc,EAAE,oBAAoB,GAAG,IAAI,CAAC;IAE5C,iDAAiD;IACjD,UAAU,EAAE,UAAU,CAAC;CACxB;AAED,gEAAgE;AAChE,wBAAgB,eAAe,CAAC,MAAM,EAAE,WAAW,GAAG,MAAM,EAAE,CAE7D;AAED,sCAAsC;AACtC,wBAAgB,oBAAoB,CAAC,MAAM,EAAE,WAAW,GAAG,MAAM,CAEhE;AAED,0DAA0D;AAC1D,wBAAgB,wBAAwB,CAAC,MAAM,EAAE,WAAW,GAAG,MAAM,CAEpE;AAED,2BAA2B;AAC3B,MAAM,WAAW,YAAY;IAC3B,2DAA2D;IAC3D,UAAU,CAAC,EAAE,UAAU,CAAC;IACxB,sCAAsC;IACtC,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,+CAA+C;IAC/C,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,iDAAiD;IACjD,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,kEAAkE;IAClE,aAAa,CAAC,EAAE,MAAM,CAAC;CACxB;AAED;;;;;;GAMG;AACH,qBAAa,MAAM;IACjB,OAAO,CAAC,OAAO,CAAkB;IACjC,OAAO,CAAC,SAAS,CAAgB;IACjC,OAAO,CAAC,kBAAkB,CAAoB;IAC9C,OAAO,CAAC,cAAc,CAAgB;IACtC,OAAO,CAAC,kBAAkB,CAA+B;IACzD,OAAO,CAAC,eAAe,CAA+B;IACtD,OAAO,CAAC,WAAW,CAAoC;gBAE3C,IAAI,CAAC,EAAE;QACjB,MAAM,CAAC,EAAE,OAAO,CAAC,eAAe,CAAC,CAAC;QAClC,QAAQ,CAAC,EAAE,aAAa,CAAC;QACzB,iBAAiB,CAAC,EAAE,iBAAiB,CAAC;QACtC,iBAAiB,CAAC,EAAE,qBAAqB,CAAC;KAC3C;IAiBD;;;;;;;OAOG;IACH,OAAO,CAAC,qBAAqB;IAiB7B,6DAA6D;IAC7D,OAAO,CAAC,iBAAiB;IAoBzB;;;;;;;;;;OAUG;IACG,KAAK,CAAC,MAAM,EAAE,MAAM,EAAE,IAAI,CAAC,EAAE,YAAY,GAAG,OAAO,CAAC,WAAW,CAAC;IAOtE;;;;;;;;OAQG;IACH,SAAS,CAAC,MAAM,EAAE,MAAM,EAAE,IAAI,CAAC,EAAE,YAAY,GAAG,WAAW;IAc3D;;;;;;OAMG;IACH,OAAO,CAAC,MAAM,CAAC,gBAAgB;IAU/B,iFAAiF;IACjF,OAAO,CAAC,YAAY;IA+CpB;;;;OAIG;IACH,QAAQ,CAAC,IAAI,EAAE;QACb,OAAO,EAAE,MAAM,CAAC;QAChB,QAAQ,EAAE,MAAM,CAAC;QACjB,UAAU,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;QACpC,OAAO,CAAC,EAAE,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;QAC3B,OAAO,CAAC,EAAE,WAAW,CAAC;QACtB,YAAY,CAAC,EAAE,MAAM,EAAE,CAAC;QACxB,WAAW,CAAC,EAAE,MAAM,CAAC;KACtB,GAAG,SAAS;IAIb;;;;;;;;;;;;;;;;OAgBG;IACG,YAAY,CAChB,IAAI,EAAE,MAAM,EACZ,OAAO,EAAE,MAAM,EAAE,EACjB,WAAW,SAAK,EAChB,QAAQ,SAAI,EACZ,QAAQ,SAAM,GACb,OAAO,CAAC,IAAI,CAAC;IAMhB;;;;;;;OAOG;IACH,gBAAgB,CACd,IAAI,EAAE,MAAM,EACZ,OAAO,EAAE,MAAM,EAAE,EACjB,WAAW,SAAK,EAChB,QAAQ,SAAI,EACZ,QAAQ,SAAM,GACb,IAAI;IAaP,+EAA+E;IAC/E,OAAO,CAAC,kBAAkB;IAsB1B,iCAAiC;IACjC,IAAI,MAAM,IAAI,aAAa,CAE1B;IAED,qCAAqC;IACrC,IAAI,UAAU,IAAI,iBAAiB,CAElC;IAED,gCAAgC;IAChC,IAAI,MAAM,IAAI,eAAe,CAE5B;CACF"}