free-coding-models 0.5.59 → 0.5.61

package/src/data/benchmarks.json

@@ -0,0 +1,302 @@
+{
+  "_meta": {
+    "schemaVersion": 1,
+    "lastUpdated": "2026-07-25",
+    "source": "Artificial Analysis + curated manual overlay",
+    "notes": "Curated seed dataset for ~30 well-known models. Refreshed by scripts/update-benchmarks.mjs before each release. Fields with null are not yet measured."
+  },
+  "deepseek-ai/deepseek-v3.2": {
+    "codingIndex": 65.4, "mathIndex": 70.2, "agenticIndex": 58.0, "reasoningIndex": 68.5,
+    "mmluPro": 75.3, "gpqa": 52.0, "hle": 8.4, "contextWindow": 160000,
+    "supportsReasoning": true, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "DeepSeek V3.2"
+  },
+  "deepseek-ai/deepseek-v4-pro": {
+    "codingIndex": 81.2, "mathIndex": 84.5, "agenticIndex": 74.8, "reasoningIndex": 80.1,
+    "mmluPro": 86.4, "gpqa": 65.7, "hle": 14.2, "contextWindow": 1000000,
+    "supportsReasoning": true, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "DeepSeek V4 Pro"
+  },
+  "deepseek-ai/deepseek-v4-flash": {
+    "codingIndex": 78.9, "mathIndex": 80.1, "agenticIndex": 71.5, "reasoningIndex": 77.0,
+    "mmluPro": 83.2, "gpqa": 60.4, "hle": 12.8, "contextWindow": 1000000,
+    "supportsReasoning": true, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "DeepSeek V4 Flash"
+  },
+  "deepseek-ai/deepseek-r1": {
+    "codingIndex": 52.0, "mathIndex": 90.4, "agenticIndex": 48.5, "reasoningIndex": 78.6,
+    "mmluPro": 84.0, "gpqa": 71.5, "hle": 9.9, "contextWindow": 128000,
+    "supportsReasoning": true, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "DeepSeek R1"
+  },
+  "deepseek/deepseek-r1": {
+    "codingIndex": 52.0, "mathIndex": 90.4, "agenticIndex": 48.5, "reasoningIndex": 78.6,
+    "mmluPro": 84.0, "gpqa": 71.5, "hle": 9.9, "contextWindow": 128000,
+    "supportsReasoning": true, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "DeepSeek R1"
+  },
+  "DeepSeek-V3.1": {
+    "codingIndex": 68.0, "mathIndex": 73.0, "agenticIndex": 60.5, "reasoningIndex": 70.0,
+    "mmluPro": 76.5, "gpqa": 53.2, "hle": 9.1, "contextWindow": 128000,
+    "supportsReasoning": true, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "DeepSeek V3.1"
+  },
+  "DeepSeek-V3.2": {
+    "codingIndex": 70.5, "mathIndex": 75.0, "agenticIndex": 62.0, "reasoningIndex": 72.0,
+    "mmluPro": 78.0, "gpqa": 54.5, "hle": 9.5, "contextWindow": 32000,
+    "supportsReasoning": true, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "DeepSeek V3.2"
+  },
+  "deepseek/deepseek-chat": {
+    "codingIndex": 65.0, "mathIndex": 70.0, "agenticIndex": 58.0, "reasoningIndex": 68.0,
+    "mmluPro": 75.0, "gpqa": 52.0, "hle": 8.4, "contextWindow": 128000,
+    "supportsReasoning": false, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "DeepSeek Chat"
+  },
+  "z-ai/glm-5.2": {
+    "codingIndex": 82.4, "mathIndex": 79.0, "agenticIndex": 76.1, "reasoningIndex": 78.9,
+    "mmluPro": 84.2, "gpqa": 61.8, "hle": 13.5, "contextWindow": 128000,
+    "supportsReasoning": true, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "GLM 5.2"
+  },
+  "zai-glm-4.7": {
+    "codingIndex": 75.0, "mathIndex": 73.5, "agenticIndex": 68.4, "reasoningIndex": 72.0,
+    "mmluPro": 78.5, "gpqa": 55.2, "hle": 10.8, "contextWindow": 128000,
+    "supportsReasoning": true, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "GLM 4.7"
+  },
+  "@cf/zai-org/glm-4.7-flash": {
+    "codingIndex": 70.0, "mathIndex": 69.0, "agenticIndex": 62.0, "reasoningIndex": 66.0,
+    "mmluPro": 73.0, "gpqa": 50.0, "hle": 9.0, "contextWindow": 128000,
+    "supportsReasoning": false, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "GLM 4.7 Flash"
+  },
+  "moonshotai/kimi-k2.6": {
+    "codingIndex": 79.5, "mathIndex": 77.0, "agenticIndex": 73.2, "reasoningIndex": 76.4,
+    "mmluPro": 82.0, "gpqa": 59.0, "hle": 12.4, "contextWindow": 262000,
+    "supportsReasoning": true, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "Kimi K2.6"
+  },
+  "@cf/moonshotai/kimi-k2.6": {
+    "codingIndex": 79.5, "mathIndex": 77.0, "agenticIndex": 73.2, "reasoningIndex": 76.4,
+    "mmluPro": 82.0, "gpqa": 59.0, "hle": 12.4, "contextWindow": 262000,
+    "supportsReasoning": true, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "Kimi K2.6"
+  },
+  "@cf/moonshotai/kimi-k2.7-code": {
+    "codingIndex": 81.0, "mathIndex": 76.5, "agenticIndex": 74.0, "reasoningIndex": 75.5,
+    "mmluPro": 81.5, "gpqa": 58.0, "hle": 12.0, "contextWindow": 262000,
+    "supportsReasoning": false, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "Kimi K2.7 Code"
+  },
+  "minimaxai/minimax-m2.7": {
+    "codingIndex": 77.5, "mathIndex": 75.0, "agenticIndex": 70.0, "reasoningIndex": 74.5,
+    "mmluPro": 80.5, "gpqa": 56.0, "hle": 11.5, "contextWindow": 200000,
+    "supportsReasoning": true, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "MiniMax M2.7"
+  },
+  "minimaxai/minimax-m3": {
+    "codingIndex": 78.0, "mathIndex": 76.0, "agenticIndex": 71.0, "reasoningIndex": 75.0,
+    "mmluPro": 81.0, "gpqa": 57.0, "hle": 12.0, "contextWindow": 1000000,
+    "supportsReasoning": true, "supportsVision": true,
+    "lastUpdated": "2026-07-20", "originalModel": "MiniMax M3"
+  },
+  "MiniMax-M2.7": {
+    "codingIndex": 77.5, "mathIndex": 75.0, "agenticIndex": 70.0, "reasoningIndex": 74.5,
+    "mmluPro": 80.5, "gpqa": 56.0, "hle": 11.5, "contextWindow": 192000,
+    "supportsReasoning": true, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "MiniMax M2.7"
+  },
+  "stepfun-ai/step-3.7-flash": {
+    "codingIndex": 73.5, "mathIndex": 74.0, "agenticIndex": 68.0, "reasoningIndex": 72.0,
+    "mmluPro": 78.0, "gpqa": 53.0, "hle": 10.5, "contextWindow": 256000,
+    "supportsReasoning": true, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "Step 3.7 Flash"
+  },
+  "openai/gpt-oss-120b": {
+    "codingIndex": 62.0, "mathIndex": 64.0, "agenticIndex": 58.0, "reasoningIndex": 63.0,
+    "mmluPro": 70.0, "gpqa": 52.0, "hle": 9.0, "contextWindow": 128000,
+    "supportsReasoning": true, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "GPT OSS 120B"
+  },
+  "gpt-oss-120b": {
+    "codingIndex": 62.0, "mathIndex": 64.0, "agenticIndex": 58.0, "reasoningIndex": 63.0,
+    "mmluPro": 70.0, "gpqa": 52.0, "hle": 9.0, "contextWindow": 128000,
+    "supportsReasoning": true, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "GPT OSS 120B"
+  },
+  "openai/gpt-oss-20b": {
+    "codingIndex": 50.0, "mathIndex": 52.0, "agenticIndex": 46.0, "reasoningIndex": 51.0,
+    "mmluPro": 58.0, "gpqa": 42.0, "hle": 6.5, "contextWindow": 128000,
+    "supportsReasoning": true, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "GPT OSS 20B"
+  },
+  "gpt-oss-20b": {
+    "codingIndex": 50.0, "mathIndex": 52.0, "agenticIndex": 46.0, "reasoningIndex": 51.0,
+    "mmluPro": 58.0, "gpqa": 42.0, "hle": 6.5, "contextWindow": 128000,
+    "supportsReasoning": true, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "GPT OSS 20B"
+  },
+  "meta/llama-4-maverick-17b-128e-instruct": {
+    "codingIndex": 73.5, "mathIndex": 70.0, "agenticIndex": 65.0, "reasoningIndex": 71.0,
+    "mmluPro": 78.0, "gpqa": 53.0, "hle": 10.0, "contextWindow": 1000000,
+    "supportsReasoning": false, "supportsVision": true,
+    "lastUpdated": "2026-07-20", "originalModel": "Llama 4 Maverick"
+  },
+  "meta-llama/llama-4-scout-17b-16e-instruct": {
+    "codingIndex": 28.0, "mathIndex": 25.0, "agenticIndex": 22.0, "reasoningIndex": 26.0,
+    "mmluPro": 30.0, "gpqa": 18.0, "hle": 2.0, "contextWindow": 10000000,
+    "supportsReasoning": false, "supportsVision": true,
+    "lastUpdated": "2026-07-20", "originalModel": "Llama 4 Scout"
+  },
+  "@cf/meta/llama-4-scout-17b-16e-instruct": {
+    "codingIndex": 28.0, "mathIndex": 25.0, "agenticIndex": 22.0, "reasoningIndex": 26.0,
+    "mmluPro": 30.0, "gpqa": 18.0, "hle": 2.0, "contextWindow": 10000000,
+    "supportsReasoning": false, "supportsVision": true,
+    "lastUpdated": "2026-07-20", "originalModel": "Llama 4 Scout"
+  },
+  "meta/llama-3.3-70b-versatile": {
+    "codingIndex": 22.0, "mathIndex": 24.0, "agenticIndex": 18.0, "reasoningIndex": 23.0,
+    "mmluPro": 28.0, "gpqa": 16.0, "hle": 1.5, "contextWindow": 131072,
+    "supportsReasoning": false, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "Llama 3.3 70B Versatile"
+  },
+  "llama-3.3-70b-versatile": {
+    "codingIndex": 22.0, "mathIndex": 24.0, "agenticIndex": 18.0, "reasoningIndex": 23.0,
+    "mmluPro": 28.0, "gpqa": 16.0, "hle": 1.5, "contextWindow": 131072,
+    "supportsReasoning": false, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "Llama 3.3 70B Versatile"
+  },
+  "Meta-Llama-3.3-70B-Instruct": {
+    "codingIndex": 22.0, "mathIndex": 24.0, "agenticIndex": 18.0, "reasoningIndex": 23.0,
+    "mmluPro": 28.0, "gpqa": 16.0, "hle": 1.5, "contextWindow": 128000,
+    "supportsReasoning": false, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "Llama 3.3 70B Instruct"
+  },
+  "llama-3.1-8b-instant": {
+    "codingIndex": 18.0, "mathIndex": 19.0, "agenticIndex": 14.0, "reasoningIndex": 17.0,
+    "mmluPro": 22.0, "gpqa": 12.0, "hle": 1.0, "contextWindow": 131072,
+    "supportsReasoning": false, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "Llama 3.1 8B Instant"
+  },
+  "qwen/qwen3.6-27b": {
+    "codingIndex": 76.5, "mathIndex": 74.0, "agenticIndex": 70.0, "reasoningIndex": 73.0,
+    "mmluPro": 79.0, "gpqa": 55.0, "hle": 11.0, "contextWindow": 131072,
+    "supportsReasoning": true, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "Qwen3.6 27B"
+  },
+  "qwen/qwen2.5-coder-32b-instruct": {
+    "codingIndex": 48.0, "mathIndex": 45.0, "agenticIndex": 42.0, "reasoningIndex": 44.0,
+    "mmluPro": 52.0, "gpqa": 32.0, "hle": 4.0, "contextWindow": 128000,
+    "supportsReasoning": false, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "Qwen2.5 Coder 32B"
+  },
+  "qwen/qwen3-30b-a3b-fp8": {
+    "codingIndex": 65.0, "mathIndex": 64.0, "agenticIndex": 58.0, "reasoningIndex": 63.0,
+    "mmluPro": 70.0, "gpqa": 48.0, "hle": 8.0, "contextWindow": 131072,
+    "supportsReasoning": true, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "Qwen3 30B A3B"
+  },
+  "mistralai/mistral-medium-3.5-128b": {
+    "codingIndex": 76.5, "mathIndex": 74.0, "agenticIndex": 70.5, "reasoningIndex": 73.0,
+    "mmluPro": 79.0, "gpqa": 56.0, "hle": 11.2, "contextWindow": 256000,
+    "supportsReasoning": false, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "Mistral Medium 3.5"
+  },
+  "mistralai/mistral-small-4-119b-2603": {
+    "codingIndex": 60.0, "mathIndex": 58.0, "agenticIndex": 55.0, "reasoningIndex": 57.0,
+    "mmluPro": 65.0, "gpqa": 44.0, "hle": 7.0, "contextWindow": 256000,
+    "supportsReasoning": false, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "Mistral Small 4"
+  },
+  "mistralai/mistral-large-3-675b-instruct-2512": {
+    "codingIndex": 58.0, "mathIndex": 56.0, "agenticIndex": 52.0, "reasoningIndex": 55.0,
+    "mmluPro": 63.0, "gpqa": 42.0, "hle": 6.5, "contextWindow": 256000,
+    "supportsReasoning": false, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "Mistral Large 675B"
+  },
+  "mistralai/ministral-14b-instruct-2512": {
+    "codingIndex": 34.0, "mathIndex": 32.0, "agenticIndex": 28.0, "reasoningIndex": 31.0,
+    "mmluPro": 38.0, "gpqa": 22.0, "hle": 3.0, "contextWindow": 32000,
+    "supportsReasoning": false, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "Ministral 14B"
+  },
+  "google/gemma-4-31b-it": {
+    "codingIndex": 52.0, "mathIndex": 50.0, "agenticIndex": 46.0, "reasoningIndex": 49.0,
+    "mmluPro": 58.0, "gpqa": 38.0, "hle": 5.5, "contextWindow": 256000,
+    "supportsReasoning": false, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "Gemma 4 31B"
+  },
+  "gemma-4-31b": {
+    "codingIndex": 52.0, "mathIndex": 50.0, "agenticIndex": 46.0, "reasoningIndex": 49.0,
+    "mmluPro": 58.0, "gpqa": 38.0, "hle": 5.5, "contextWindow": 128000,
+    "supportsReasoning": false, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "Gemma 4 31B"
+  },
+  "@cf/google/gemma-4-26b-a4b-it": {
+    "codingIndex": 48.0, "mathIndex": 46.0, "agenticIndex": 42.0, "reasoningIndex": 45.0,
+    "mmluPro": 54.0, "gpqa": 35.0, "hle": 5.0, "contextWindow": 128000,
+    "supportsReasoning": false, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "Gemma 4 26B"
+  },
+  "nvidia/nemotron-3-ultra-550b-a55b": {
+    "codingIndex": 71.0, "mathIndex": 70.0, "agenticIndex": 65.0, "reasoningIndex": 69.0,
+    "mmluPro": 75.0, "gpqa": 52.0, "hle": 9.5, "contextWindow": 1000000,
+    "supportsReasoning": true, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "Nemotron 3 Ultra"
+  },
+  "nvidia/nemotron-3-super-120b-a12b": {
+    "codingIndex": 60.0, "mathIndex": 59.0, "agenticIndex": 54.0, "reasoningIndex": 58.0,
+    "mmluPro": 65.0, "gpqa": 44.0, "hle": 7.0, "contextWindow": 128000,
+    "supportsReasoning": true, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "Nemotron 3 Super"
+  },
+  "nvidia/nemotron-3-nano-30b-a3b": {
+    "codingIndex": 38.0, "mathIndex": 36.0, "agenticIndex": 32.0, "reasoningIndex": 35.0,
+    "mmluPro": 42.0, "gpqa": 26.0, "hle": 3.5, "contextWindow": 1000000,
+    "supportsReasoning": false, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "Nemotron 3 Nano"
+  },
+  "bytedance/seed-oss-36b-instruct": {
+    "codingIndex": 56.0, "mathIndex": 54.0, "agenticIndex": 50.0, "reasoningIndex": 53.0,
+    "mmluPro": 60.0, "gpqa": 40.0, "hle": 6.0, "contextWindow": 32000,
+    "supportsReasoning": false, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "Seed OSS 36B"
+  },
+  "stockmark/stockmark-2-100b-instruct": {
+    "codingIndex": 36.0, "mathIndex": 34.0, "agenticIndex": 30.0, "reasoningIndex": 33.0,
+    "mmluPro": 40.0, "gpqa": 24.0, "hle": 3.0, "contextWindow": 32000,
+    "supportsReasoning": false, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "Stockmark 100B"
+  },
+  "codestral-latest": {
+    "codingIndex": 42.0, "mathIndex": 40.0, "agenticIndex": 38.0, "reasoningIndex": 39.0,
+    "mmluPro": 48.0, "gpqa": 30.0, "hle": 4.0, "contextWindow": 32000,
+    "supportsReasoning": false, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "Codestral Latest"
+  },
+  "qwen/qwq-32b": {
+    "codingIndex": 58.0, "mathIndex": 82.0, "agenticIndex": 50.0, "reasoningIndex": 76.0,
+    "mmluPro": 75.0, "gpqa": 64.0, "hle": 11.0, "contextWindow": 131072,
+    "supportsReasoning": true, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "QwQ 32B"
+  },
+  "poolside/laguna-m.1:free": {
+    "codingIndex": 72.0, "mathIndex": 68.0, "agenticIndex": 65.0, "reasoningIndex": 69.0,
+    "mmluPro": 74.0, "gpqa": 50.0, "hle": 9.5, "contextWindow": 262000,
+    "supportsReasoning": false, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "Poolside Laguna M.1"
+  },
+  "poolside/laguna-xs.2:free": {
+    "codingIndex": 68.0, "mathIndex": 65.0, "agenticIndex": 62.0, "reasoningIndex": 65.0,
+    "mmluPro": 70.0, "gpqa": 47.0, "hle": 8.5, "contextWindow": 262000,
+    "supportsReasoning": false, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "Poolside Laguna XS.2"
+  },
+  "@cf/qwen/qwen3-30b-a3b-fp8": {
+    "codingIndex": 65.0, "mathIndex": 64.0, "agenticIndex": 58.0, "reasoningIndex": 63.0,
+    "mmluPro": 70.0, "gpqa": 48.0, "hle": 8.0, "contextWindow": 131072,
+    "supportsReasoning": true, "supportsVision": false,
+    "lastUpdated": "2026-07-20", "originalModel": "Qwen3 30B A3B"
+  }
+}

package/src/tui/app.js

@@ -149,6 +149,69 @@ const mergedModels = buildMergedModels(MODELS)
 const mergedModelByLabel = new Map(mergedModels.map(m => [m.label, m]))
 setOpenCodeModelData(mergedModels, mergedModelByLabel)
 
+// 📖 Enrichment (t4 + t5): module-level state. The sync part (extended
+// 📖 benchmark catalog stats) is computed lazily on first read via the
+// 📖 getter, so we don't block module load. The async part (models.dev
+// 📖 fetch) runs inside runApp as a background task — never at module load,
+// 📖 never blocking the TUI from rendering.
+// 📖
+// 📖 The TUI reads from this object on every render to update the footer chips.
+const moduleEnrichmentStats = {
+  bench: { total: 0, lastUpdated: 'unknown', source: 'unknown' },
+  modelsDev: { 'sources.js': 0, 'models.dev': 0, cached: false },
+}
+let _benchStatsLoaded = false
+function ensureBenchStatsLoaded() {
+  if (_benchStatsLoaded) return
+  _benchStatsLoaded = true
+  try {
+    // 📖 require() works synchronously and is safe at module load because
+    // 📖 model-merger.js has no top-level await side effects.
+    const { getEnrichmentStats } = require('../core/model-merger.js')
+    moduleEnrichmentStats.bench = getEnrichmentStats().extendedBench
+  } catch (err) {
+    if (process.env.FCM_BENCH_DEBUG) console.warn('[t4] enrichment stats failed:', err?.message ?? err)
+  }
+}
+
+// 📖 Background enrichment: fetch models.dev and update moduleEnrichmentStats
+// 📖 when it resolves. Fully decoupled from runApp — even if the fetch hangs
+// 📖 for 30s+, the TUI keeps rendering. The chip is updated by the next render
+// 📖 after resolution.
+function runModelsDevEnrichmentInBackground(merged) {
+  // 📖 Fire-and-forget; never awaited. If it throws, the catch swallows it.
+  ;(async () => {
+    try {
+      // 📖 Bound the total wait to 12s (worst case 3×4s retries). If the
+      // 📖 network is truly down, the chip stays at "0 live" and the TUI
+      // 📖 is unaffected. Callers can still trigger a refresh via the
+      // 📖 settings panel (future work).
+      const timeout = new Promise(resolve => setTimeout(() => resolve(null), 12_000))
+      const { overlayModelsDevMetadata } = await import('../core/model-merger.js')
+      const result = await Promise.race([
+        overlayModelsDevMetadata(merged, { mutate: true }),
+        timeout,
+      ])
+      if (!result) {
+        if (process.env.FCM_BENCH_DEBUG) console.warn('[t5] models.dev fetch timed out (12s)')
+        return
+      }
+      let live = 0, curated = 0
+      for (const m of result) {
+        if (m.metaSource === 'models.dev') live++
+        else if (m.metaSource === 'sources.js') curated++
+      }
+      moduleEnrichmentStats.modelsDev = { 'sources.js': curated, 'models.dev': live, cached: true }
+      // 📖 Note: state is owned by runApp — we don't have a direct ref here.
+      // 📖 The footer reads moduleEnrichmentStats lazily on each render, so
+      // 📖 the next render will pick up the new counts. We do NOT trigger a
+      // 📖 render from here (we're not in the render context).
+    } catch (err) {
+      if (process.env.FCM_BENCH_DEBUG) console.warn('[t5] models.dev overlay failed:', err?.message ?? err)
+    }
+  })()
+}
+
 // 📖 Provider quota cache is managed by lib/provider-quota-fetchers.js (TTL + backoff).
 // 📖 Usage placeholder logic uses isKnownQuotaTelemetry() from lib/quota-capabilities.js.
 
@@ -189,6 +252,16 @@ const LOCAL_VERSION = pkg.version
 // 📖 OpenCode helpers are imported from ../src/opencode.js
 
 export async function runApp(cliArgs, config, startupOptions = {}) {
+  // 📖 t4: prime the sync extended-benchmark stats so the footer chip is
+  // 📖 accurate on the first render (no async, no network — just a sync read).
+  ensureBenchStatsLoaded()
+
+  // 📖 t5: kick off the async models.dev enrichment in the background. This
+  // 📖 does NOT block runApp — the TUI starts rendering immediately with
+  // 📖 curated values, then re-renders when the models.dev fetch resolves.
+  // 📖 If the network is unreachable, the chip just stays at "0 live" forever
+  // 📖 (no user-visible freeze). The background task is bounded to 12s.
+  runModelsDevEnrichmentInBackground(mergedModels)
 
   // 📖 Detect user active terminal theme
   detectActiveTheme(config.settings?.theme || 'auto')
@@ -505,8 +578,27 @@ export async function runApp(cliArgs, config, startupOptions = {}) {
   }
   state.probeCacheHits = probeCacheHits
   state.probeCacheMisses = probeCacheMisses
+  // 📖 Enrichment (t4 + t5): pulled fresh on every render from moduleEnrichmentStats
+  // 📖 (see the tableOpts construction below). No need to mirror into state here.
   state.probeCacheBrokenHidden = state.results.filter(r => r.cachedBroken && r.hidden).length
 
+  // 📖 Runtime telemetry (t3): load the per-model metrics file + compute the
+  // 📖 real-world score for each result so the table can sort/display it.
+  // 📖 The file may not exist yet (no daemon traffic) — that's fine, every
+  // 📖 r.realWorldScore stays null and we fall back to SWE-bench.
+  try {
+    const { loadRuntimeTelemetry, getRealWorldScore } = await import('../core/runtime-telemetry.js')
+    loadRuntimeTelemetry()  // 📖 populate module-level cache
+    for (const r of state.results) {
+      const score = getRealWorldScore(r.providerKey, r.modelId)
+      r.realWorldScore = score  // null when below MIN_CALLS_FOR_SCORE
+    }
+    state.runtimeTelemetryLoaded = true
+  } catch (err) {
+    state.runtimeTelemetryLoaded = false
+    state.runtimeTelemetryError = err?.message || String(err)
+  }
+
   // 📖 Define pingModel before JSON mode so `--json` can reuse the same provider-aware
   // 📖 ping path as the interactive TUI without waiting for the PTY/render loop setup.
   pingModel = async (r) => {
@@ -954,6 +1046,17 @@ export async function runApp(cliArgs, config, startupOptions = {}) {
       probeCacheMisses: state.probeCacheMisses || 0,
       probeCacheBrokenHidden: state.probeCacheBrokenHidden || 0,
       showBrokenMode: !!state.showBrokenMode,
+      // 📖 Enrichment chips (t4 + t5): benchmark catalog + models.dev provenance
+      // 📖 Read fresh from moduleEnrichmentStats on every render so the async
+      // 📖 models.dev fetch (background, never blocks the TUI) is reflected as
+      // 📖 soon as it resolves — even if the user isn't pressing keys.
+      enrichmentBenchCount: moduleEnrichmentStats.bench.total || 0,
+      enrichmentBenchLastUpdated: moduleEnrichmentStats.bench.lastUpdated || null,
+      metaSourceCounts: {
+        'sources.js': moduleEnrichmentStats.modelsDev['sources.js'] || 0,
+        'models.dev': moduleEnrichmentStats.modelsDev['models.dev'] || 0,
+      },
+      modelsDevCacheCached: moduleEnrichmentStats.modelsDev.cached === true,
       // 📖 Passive quota (t2): live snapshots from response headers, populated
       // 📖 by pings (every ping pre-warms quota for its provider). See
       // 📖 src/core/provider-quota-fetchers.js for getAllPassiveQuotas().

package/src/tui/cli-help.js

@@ -36,6 +36,8 @@ const ANALYSIS_FLAGS = [
   { flag: '--reprobe, --no-cache', description: 'Force-rebuild the persistent probe-cache this run (skip cached healthy models)' },
   { flag: '--probe-ttl <ms>', description: 'Override probe-cache TTL (default 24h); healthy models within TTL are not re-pinged' },
   { flag: '--show-broken', description: "Don't auto-hide models that the probe-cache marked broken (one-shot override)" },
+  { flag: '--check-drift', description: 'Diff sources.js against models.dev and print a drift report; exit 1 on mismatch (t5)' },
+  { flag: '--drift-threshold <N>', description: 'Only fail --check-drift when N+ mismatches are found (default: 0 = any drift)' },
 ]
 
 const CONFIG_FLAGS = [

package/src/tui/key-handler.js

@@ -836,6 +836,40 @@ export function createKeyHandler(ctx) {
     state.tokenUsageError = null
   }
 
+  // 📖 Runtime Report overlay (t3): per-model real-world breakdown + recent calls.
+  // 📖 Reads the local ~/.free-coding-models/runtime-telemetry.json file directly
+  // 📖 (the daemon writes to the same file; the CLI doesn't need to talk to it).
+  async function openRuntimeReportOverlay() {
+    state.runtimeReportOpen = true
+    state.runtimeReportScrollOffset = 0
+    state.runtimeReportError = null
+    state.runtimeReportData = null
+    state.runtimeReportSelectedKey = null
+    try {
+      const { loadRuntimeTelemetry, getAllModelTelemetry } = await import('../core/runtime-telemetry.js')
+      const cache = loadRuntimeTelemetry()
+      const all = getAllModelTelemetry({ cache })
+      const entries = Object.entries(all).sort((a, b) => (b[1].totalCalls || 0) - (a[1].totalCalls || 0))
+      state.runtimeReportData = entries
+      // 📖 Pre-select the currently-focused model if it has telemetry.
+      const focused = state.visibleSorted?.[state.cursor]
+      if (focused) {
+        const key = `${focused.providerKey}/${focused.modelId}`
+        if (entries.find(([k]) => k === key)) state.runtimeReportSelectedKey = key
+      }
+    } catch (err) {
+      state.runtimeReportError = err?.message || 'Failed to load runtime telemetry'
+    }
+  }
+
+  function closeRuntimeReportOverlay() {
+    state.runtimeReportOpen = false
+    state.runtimeReportScrollOffset = 0
+    state.runtimeReportError = null
+    state.runtimeReportData = null
+    state.runtimeReportSelectedKey = null
+  }
+
 
   function cycleToolMode() {
     const modeOrder = getToolModeOrder()
@@ -1355,6 +1389,7 @@ export function createKeyHandler(ctx) {
       case 'sort-verdict': return setSortColumnFromCommand('verdict')
       case 'sort-stability': return setSortColumnFromCommand('stability')
       case 'sort-uptime': return setSortColumnFromCommand('uptime')
+      case 'sort-realworld': return setSortColumnFromCommand('realworld')
       case 'open-settings': return openSettingsOverlay()
       case 'open-help':
         state.helpVisible = true
@@ -1366,6 +1401,7 @@ export function createKeyHandler(ctx) {
       case 'open-router-dashboard': return openRouterDashboardOverlay(state)
       case 'open-playground': return openPlaygroundOverlay(state)
       case 'open-token-usage': return openTokenUsageOverlay()
+      case 'open-runtime-report': return openRuntimeReportOverlay()
       case 'open-install-endpoints': return openInstallEndpointsOverlay()
       case 'open-installed-models': return openInstalledModelsOverlay()
       case 'action-cycle-theme': return cycleGlobalTheme()
@@ -2120,6 +2156,42 @@ export function createKeyHandler(ctx) {
       return
     }
 
+    // 📖 Runtime Report overlay (t3): Shift+W. Per-model real-world breakdown +
+    // 📖 recent calls. Scrolling with up/down/pageup/pagedown/escape.
+    if (state.runtimeReportOpen) {
+      if (key.ctrl && key.name === 'c') { exit(0); return }
+      const pageStep = Math.max(1, (state.terminalRows || 1) - 4)
+      if (key.name === 'escape') {
+        closeRuntimeReportOverlay()
+        return
+      }
+      if (key.name === 'up' || key.name === 'k') {
+        state.runtimeReportScrollOffset = Math.max(0, state.runtimeReportScrollOffset - 1)
+        return
+      }
+      if (key.name === 'down' || key.name === 'j') {
+        state.runtimeReportScrollOffset += 1
+        return
+      }
+      if (key.name === 'pageup') {
+        state.runtimeReportScrollOffset = Math.max(0, state.runtimeReportScrollOffset - pageStep)
+        return
+      }
+      if (key.name === 'pagedown') {
+        state.runtimeReportScrollOffset += pageStep
+        return
+      }
+      if (key.name === 'home') {
+        state.runtimeReportScrollOffset = 0
+        return
+      }
+      if (key.name === 'end') {
+        state.runtimeReportScrollOffset = Number.MAX_SAFE_INTEGER
+        return
+      }
+      return
+    }
+
     // 📖 Router Onboarding overlay: shown on first launch. Y=yes enable, N=not now, Esc=cancel.
     if (state.routerOnboardingOpen) {
       if (key.ctrl && key.name === 'c') { exit(0); return }
@@ -2872,6 +2944,14 @@ export function createKeyHandler(ctx) {
       return
     }
 
+    // 📖 Shift+W: open the Runtime Report overlay (t3) — per-model real-world
+    // 📖 success rate + throughput + recent calls. See command palette
+    // 📖 'open-runtime-report' for the same action.
+    if (key.name === 'w' && key.shift && !key.alt && !key.ctrl && !key.meta) {
+      openRuntimeReportOverlay()
+      return
+    }
+
     // 📖 E cycles: Normal → Configured only → Usable only → Normal
     // 📖 Configured only: hides models with no key or noauth/auth_error health
     // 📖 Usable only: only shows models with Health UP and Verdict ≤ Slow (Perfect/Normal/Slow)

package/src/tui/render-table.js

@@ -155,6 +155,13 @@ export const PROVIDER_COLOR = new Proxy({}, {
  *   probeCacheMisses?: number,       // t1: number of models that needed a live probe
  *   probeCacheBrokenHidden?: number, // t1: number of broken models auto-hidden this session
  *   showBrokenMode?: boolean,        // t1: true when Shift+B has un-hidden broken models
+ *   enrichmentBenchCount?: number,   // t4: extended benchmark catalog size (committed JSON)
+ *   enrichmentBenchLastUpdated?: string, // t4: lastUpdated field from the catalog _meta
+ *   metaSourceCounts?: {             // t5: count of models per metadata source
+ *     'sources.js': number,
+ *     'models.dev': number,
+ *   },
+ *   modelsDevCacheCached?: boolean,  // t5: whether the models.dev cache is still within TTL
  *   quota?: Record<string, {         // t2: live quota from response headers
  *     remaining: number, limit: number, percent: number,
  *     source: 'header'|'endpoint', lastUpdated: number, windowType?: string,
@@ -209,6 +216,10 @@ export function renderTable({
   probeCacheMisses = 0,
   probeCacheBrokenHidden = 0,
   showBrokenMode = false,
+  enrichmentBenchCount = 0,
+  enrichmentBenchLastUpdated = null,
+  metaSourceCounts = null,
+  modelsDevCacheCached = false,
   quota = {},
 } = _) {
   // 📖 Filter out hidden models for display
@@ -1224,6 +1235,29 @@ export function renderTable({
     probeCacheLabel = chalk.bgRgb(40, 90, 140).rgb(220, 235, 255).bold(` ${cachedTxt}${brokenTxt} `)
   }
 
+  // 📖 Enrichment chips (t4 + t5):
+  // 📖 `📊 bench N (date) · 📡 live K / 📦 cached M`
+  // 📖 Always shown when any enrichment state exists.
+  let enrichmentLabel = ''
+  if ((enrichmentBenchCount && enrichmentBenchCount > 0) || metaSourceCounts) {
+    const parts = []
+    if (enrichmentBenchCount && enrichmentBenchCount > 0) {
+      const dateTxt = enrichmentBenchLastUpdated && enrichmentBenchLastUpdated !== 'unknown'
+        ? ` (${enrichmentBenchLastUpdated})`
+        : ''
+      parts.push(`📊 bench ${enrichmentBenchCount}${dateTxt}`)
+    }
+    if (metaSourceCounts && (metaSourceCounts['models.dev'] > 0 || metaSourceCounts['sources.js'] > 0)) {
+      const liveTxt = metaSourceCounts['models.dev'] ?? 0
+      const cachedTxt = metaSourceCounts['sources.js'] ?? 0
+      const liveIcon = modelsDevCacheCached ? '📡' : '📦'
+      parts.push(`${liveIcon} ${liveTxt} live · ${cachedTxt} curated`)
+    }
+    if (parts.length > 0) {
+      enrichmentLabel = chalk.bgRgb(80, 50, 110).rgb(230, 220, 245).bold(` ${parts.join(' · ')} `)
+    }
+  }
+
   // 📖 Passive quota chip (t2): `📊 groq 78% · sambanova 41%` (top-N depleted first).
   // 📖 Shows live rate-limit headers that the daemon / pings collected from upstream
   // 📖 responses. Zero extra network requests — see provider-quota-fetchers.js.
@@ -1243,8 +1277,8 @@ export function renderTable({
     }
   }
 
-  // 📖 Line 3: Speed Test + Global Benchmark + Probe + Probe-cache + Quota + Last release
-  if (releaseLabel || speedTestLabel || globalBenchmarkLabel || probeLabel || probeCacheLabel || quotaLabel) {
+  // 📖 Line 3: Speed Test + Global Benchmark + Probe + Probe-cache + Enrichment + Quota + Last release
+  if (releaseLabel || speedTestLabel || globalBenchmarkLabel || probeLabel || probeCacheLabel || enrichmentLabel || quotaLabel) {
     const parts = [
       { text: '  ', key: null },
       { text: speedTestLabel, key: 'a' },
@@ -1254,6 +1288,8 @@ export function renderTable({
       { text: probeLabel, key: null },
       { text: probeCacheLabel ? '  ' : '', key: null },
       { text: probeCacheLabel, key: null },
+      { text: enrichmentLabel ? '  ' : '', key: null },
+      { text: enrichmentLabel, key: null },
       { text: quotaLabel ? '  ' : '', key: null },
       { text: quotaLabel, key: null },
       { text: '  ', key: null },

package/src/tui/tui-state.js

@@ -297,6 +297,15 @@ export function createTuiState({
     probeCacheTtlMs: null,        // 📖 Set in app.js from cliArgs.probeTtlMs || DEFAULT
     showBrokenMode: false,
 
+    // 📖 Runtime telemetry overlay state (t3): Shift+W opens the Runtime Report
+    // 📖 overlay. Data is loaded fresh from ~/.free-coding-models/runtime-telemetry.json
+    // 📖 when the overlay opens — no polling, no daemon dependency.
+    runtimeReportOpen: false,
+    runtimeReportScrollOffset: 0,
+    runtimeReportError: null,
+    runtimeReportData: null,        // 📖 Array<[key, ModelTelemetry]>
+    runtimeReportSelectedKey: null, // 📖 Pre-selected model key
+
     // 📖 Header click flash animation: briefly highlights the clicked column header
     // 📖 with an inverse/bright style for ~250ms (3 frames at 12 FPS).
     headerFlashColumn: null,       // 📖 Column name being flashed (null = no flash active)