@x12i/ai-profiles 1.8.0 → 2.0.1

package/dist/data/models-profiles.json

@@ -2,8 +2,8 @@
   "schema": "x12i.ai-profiles.model-registry.v3",
   "registryId": "models-profiles",
   "generatedAt": "2026-05-29T00:00:00.000Z",
-  "lastVerifiedAt": "2026-05-30T21:56:06Z",
-  "version": "3.0.0",
+  "lastVerifiedAt": "2026-06-01T12:00:00Z",
+  "version": "4.0.0",
   "currency": "USD",
   "pricingUnit": "1M_tokens",
   "namingConvention": "lowerCamelCase",
@@ -17,7 +17,12 @@
     "cpuLocalProfile": "Only the \"local\" profile runs on-device CPU (funcx / Transformers.js). No OpenRouter.",
     "inputModalities": "Required input media types for profile intent (text, image, audio, video, file).",
     "outputModalities": "Required output media types for profile intent (text, image, audio, video, file).",
-    "choiceModalities": "Optional per-choice modality override when catalog linkage is missing."
+    "choiceModalities": "Optional per-choice modality override when catalog linkage is missing.",
+    "profileTags": "Short discovery tags (2–4 chars). Vendor affinity: ds, ggl, xai, mmx, ant, oai. (profile, tag-set) pairs are unique.",
+    "catalogLane": "Required OpenRouter browser lane per profile. resolveAIProfile requires options.catalogLane matching the profile lane.",
+    "secondary": "Optional companion model per choice (review, escalation, verification). Resolved via resolveSecondaryModel; not a failover replacement.",
+    "fallbackChoiceOrder": "Replacement order when constraints exclude the primary choice (cost cap, vendor trim). Independent of secondary.",
+    "volRouting": "Volume agent profile (vol): Flash default, Pro for hard/final. Model SKUs are never profile aliases."
   },
   "sourceCatalog": {
     "providers": [
@@ -58,6 +63,15 @@
       "displayName": "Cheap",
       "description": "Low-cost model selection for simple, high-volume, non-critical tasks. This does not always mean the absolute cheapest model.",
       "category": "cost",
+      "tags": [
+        "eco",
+        "wf",
+        "xtr",
+        "cls",
+        "bat",
+        "tri",
+        "ggl"
+      ],
       "aliases": [
         "economy",
         "budget",
@@ -87,7 +101,12 @@
               "openrouter:catalog"
             ]
           },
-          "modelStatus": "verified"
+          "modelStatus": "verified",
+          "secondary": {
+            "kind": "choice",
+            "choice": "anthropic_cheap",
+            "reason": "Stronger instruction-following pass when cheap default quality is insufficient."
+          }
         },
         "google_floor": {
           "choice": "google_floor",
@@ -154,7 +173,7 @@
               "openrouter:catalog"
             ]
           },
-          "modelStatus": "predicted"
+          "modelStatus": "verified"
         },
         "minimax_cheap": {
           "choice": "minimax_cheap",
@@ -198,13 +217,22 @@
         "outputModalities": [
           "text"
         ]
-      }
+      },
+      "catalogLane": "text"
     },
     "fast": {
       "profile": "fast",
       "displayName": "Fast",
       "description": "Low-latency model selection for interactive product flows.",
       "category": "speed",
+      "tags": [
+        "spd",
+        "wf",
+        "agt",
+        "cod",
+        "ggl",
+        "xai"
+      ],
       "aliases": [
         "instant",
         "quick",
@@ -342,13 +370,21 @@
         "outputModalities": [
           "text"
         ]
-      }
+      },
+      "catalogLane": "text"
     },
     "balanced": {
       "profile": "balanced",
       "displayName": "Balanced",
       "description": "Default model selection for normal graph execution.",
       "category": "standard",
+      "tags": [
+        "wf",
+        "anl",
+        "cod",
+        "agt",
+        "ggl"
+      ],
       "aliases": [
         "default",
         "standard",
@@ -377,7 +413,12 @@
               "openrouter:catalog"
             ]
           },
-          "modelStatus": "verified"
+          "modelStatus": "verified",
+          "secondary": {
+            "kind": "choice",
+            "choice": "anthropic_balanced",
+            "reason": "Higher-quality reviewer when balanced default needs a second opinion."
+          }
         },
         "anthropic_balanced": {
           "choice": "anthropic_balanced",
@@ -462,13 +503,25 @@
         "outputModalities": [
           "text"
         ]
-      }
+      },
+      "catalogLane": "text"
     },
     "deep": {
       "profile": "deep",
       "displayName": "Deep",
       "description": "High-reasoning model selection for complex work.",
       "category": "reasoning",
+      "tags": [
+        "rsn",
+        "anl",
+        "cod",
+        "eng",
+        "exp",
+        "pro",
+        "rev",
+        "ds",
+        "xai"
+      ],
       "aliases": [
         "thinking",
         "reasoning",
@@ -498,7 +551,12 @@
               "openrouter:catalog"
             ]
           },
-          "modelStatus": "verified"
+          "modelStatus": "verified",
+          "secondary": {
+            "kind": "choice",
+            "choice": "deepseek_pro",
+            "reason": "Alternate flagship reasoning tier for verification and escalation."
+          }
         },
         "openai_deep": {
           "choice": "openai_deep",
@@ -557,13 +615,14 @@
           },
           "reason": "Flagship-class reasoning without Opus or GPT-5.5 pricing.",
           "verification": {
-            "lastVerifiedAt": "2026-05-29T17:08:03Z",
+            "lastVerifiedAt": "2026-06-01T12:00:00Z",
             "sourceRefs": [
               "vendor:deepseek",
-              "openrouter:catalog"
+              "openrouter:catalog",
+              "vendor:deepseek-api-docs"
             ]
           },
-          "modelStatus": "predicted"
+          "modelStatus": "verified"
         },
         "grok_flagship": {
           "choice": "grok_flagship",
@@ -631,13 +690,22 @@
         "outputModalities": [
           "text"
         ]
-      }
+      },
+      "catalogLane": "text"
     },
     "pro": {
       "profile": "pro",
       "displayName": "Pro",
       "description": "Premium model selection for critical tasks.",
       "category": "premium",
+      "tags": [
+        "flg",
+        "pro",
+        "exp",
+        "anl",
+        "cod",
+        "rev"
+      ],
       "aliases": [
         "premium",
         "best",
@@ -727,13 +795,21 @@
         "outputModalities": [
           "text"
         ]
-      }
+      },
+      "catalogLane": "text"
     },
     "json": {
       "profile": "json",
       "displayName": "JSON",
       "description": "Model selection for strict JSON, schema, extraction, and classification tasks.",
       "category": "output",
+      "tags": [
+        "str",
+        "json",
+        "xtr",
+        "cls",
+        "eco"
+      ],
       "aliases": [
         "schema",
         "strict-json",
@@ -768,7 +844,12 @@
               "openrouter:catalog"
             ]
           },
-          "modelStatus": "verified"
+          "modelStatus": "verified",
+          "secondary": {
+            "kind": "choice",
+            "choice": "anthropic_json",
+            "reason": "Higher-quality structured-output reviewer for complex extraction."
+          }
         },
         "google_json": {
           "choice": "google_json",
@@ -813,10 +894,35 @@
             ]
           },
           "modelStatus": "predicted"
+        },
+        "deepseek_json": {
+          "choice": "deepseek_json",
+          "displayName": "DeepSeek V4 Flash JSON",
+          "description": "Low-cost structured output without thinking overhead.",
+          "provider": "deepseek",
+          "modelId": "deepseek-v4-flash",
+          "modelStatus": "verified",
+          "runtime": {
+            "instructionTier": "default",
+            "backend": "openrouter",
+            "reasoningEffort": "minimal",
+            "toolPolicy": "none",
+            "outputMode": "schema",
+            "temperature": 0
+          },
+          "reason": "Classification and extraction at DeepSeek Flash pricing; no expensive reasoning.",
+          "verification": {
+            "lastVerifiedAt": "2026-06-01T12:00:00Z",
+            "sourceRefs": [
+              "vendor:deepseek",
+              "openrouter:catalog"
+            ]
+          }
         }
       },
       "fallbackChoiceOrder": [
         "default",
+        "deepseek_json",
         "google_json",
         "anthropic_json"
       ],
@@ -830,13 +936,19 @@
         "outputModalities": [
           "text"
         ]
-      }
+      },
+      "catalogLane": "text"
     },
     "agentic": {
       "profile": "agentic",
       "displayName": "Agentic",
       "description": "Model selection for tool use, planning, and multi-step execution.",
       "category": "agentic",
+      "tags": [
+        "agt",
+        "wf",
+        "cod"
+      ],
       "aliases": [
         "agent",
         "tools",
@@ -869,7 +981,12 @@
               "openrouter:catalog"
             ]
           },
-          "modelStatus": "verified"
+          "modelStatus": "verified",
+          "secondary": {
+            "kind": "choice",
+            "choice": "openai_agentic",
+            "reason": "Cross-vendor reviewer for agentic tool-use workflows."
+          }
         },
         "openai_agentic": {
           "choice": "openai_agentic",
@@ -939,6 +1056,7 @@
       },
       "fallbackChoiceOrder": [
         "default",
+        "deepseek_agentic",
         "openai_agentic",
         "google_agentic",
         "minimax_agentic"
@@ -954,13 +1072,19 @@
         "outputModalities": [
           "text"
         ]
-      }
+      },
+      "catalogLane": "text"
     },
     "research": {
       "profile": "research",
       "displayName": "Research",
       "description": "Model selection for source-heavy reading, synthesis, and cited research.",
       "category": "research",
+      "tags": [
+        "res",
+        "anl",
+        "rev"
+      ],
       "aliases": [
         "source-review",
         "web-research",
@@ -1064,27 +1188,65 @@
         "outputModalities": [
           "text"
         ]
-      }
+      },
+      "catalogLane": "text"
     },
     "cyber": {
       "profile": "cyber",
       "displayName": "Cyber & Risk Analysis",
-      "description": "Model selection tuned for threat analysis, TPRM, contract verification, and log parsing.",
+      "description": "Model selection tuned for threat analysis, TPRM, contract verification, log parsing, and security research.",
       "category": "security",
+      "tags": [
+        "cyb",
+        "sec",
+        "anl",
+        "for",
+        "aud",
+        "exp",
+        "eco",
+        "res",
+        "ds"
+      ],
       "aliases": [
         "security",
         "threat-intel",
         "risk",
-        "audit"
+        "audit",
+        "cyber-analysis",
+        "cybersecurity"
       ],
       "defaultChoice": "default",
       "choices": {
         "default": {
           "choice": "default",
-          "displayName": "Verification Default",
-          "description": "Balanced reasoning for policy review and contract verification.",
-          "provider": "anthropic",
-          "modelId": "claude-sonnet-4-5",
+          "displayName": "Cyber Default (DeepSeek V4 Flash)",
+          "description": "Default cyber analysis: fast, low-cost, 1M context for most TPRM, triage, and agent steps.",
+          "provider": "deepseek",
+          "modelId": "deepseek-v4-flash",
+          "runtime": {
+            "instructionTier": "default",
+            "backend": "openrouter",
+            "reasoningEffort": "medium",
+            "toolPolicy": "read-only",
+            "outputMode": "text"
+          },
+          "reason": "Best default for high-volume cyber workflows; escalate to Pro or Grok for harder forensics.",
+          "verification": {
+            "lastVerifiedAt": "2026-06-01T12:00:00Z",
+            "sourceRefs": [
+              "vendor:deepseek",
+              "openrouter:catalog",
+              "vendor:deepseek-api-docs"
+            ]
+          },
+          "modelStatus": "verified"
+        },
+        "grok_intel": {
+          "choice": "grok_intel",
+          "displayName": "Grok 4.3 Intel",
+          "description": "Real-time threat intel and large-context log review (1M tokens).",
+          "provider": "xai",
+          "modelId": "grok-4.3",
           "runtime": {
             "instructionTier": "reasoning",
             "backend": "openrouter",
@@ -1092,11 +1254,11 @@
             "toolPolicy": "read-only",
             "outputMode": "text"
           },
-          "reason": "Nuanced reasoning for loopholes, fraud indicators, and subtle contract risks.",
+          "reason": "Strong for bulk log ingestion, MCP workflows, and comparative cyber research vs GPT-5.5.",
           "verification": {
-            "lastVerifiedAt": "2026-05-29T17:08:03Z",
+            "lastVerifiedAt": "2026-06-01T12:00:00Z",
             "sourceRefs": [
-              "vendor:anthropic",
+              "vendor:xai",
               "openrouter:catalog"
             ]
           },
@@ -1105,7 +1267,7 @@
         "log_analysis": {
           "choice": "log_analysis",
           "displayName": "Bulk Log Parser",
-          "description": "High-volume, large-context analysis for server logs or network traffic.",
+          "description": "High-volume structured parsing of server logs or network traffic.",
           "provider": "google",
           "modelId": "gemini-3.5-flash",
           "runtime": {
@@ -1125,8 +1287,31 @@
         },
         "deep_forensics": {
           "choice": "deep_forensics",
-          "displayName": "Deep Forensics",
-          "description": "Maximum reasoning for vulnerability hunting and code auditing.",
+          "displayName": "Deep Forensics (DeepSeek V4 Pro)",
+          "description": "Hard reasoning for vulnerability hunting, exploit chaining, and code auditing.",
+          "provider": "deepseek",
+          "modelId": "deepseek-v4-pro",
+          "runtime": {
+            "instructionTier": "reasoning",
+            "backend": "openrouter",
+            "reasoningEffort": "high",
+            "toolPolicy": "read-only",
+            "outputMode": "text"
+          },
+          "reason": "Economic pro-tier forensics; use gpt_forensics when you need OpenAI Trusted Access validation paths.",
+          "verification": {
+            "lastVerifiedAt": "2026-06-01T12:00:00Z",
+            "sourceRefs": [
+              "vendor:deepseek",
+              "openrouter:catalog"
+            ]
+          },
+          "modelStatus": "verified"
+        },
+        "gpt_forensics": {
+          "choice": "gpt_forensics",
+          "displayName": "GPT-5.5 Forensics (flagship)",
+          "description": "Maximum validated reasoning for zero-day research and complex attack paths.",
           "provider": "openai",
           "modelId": "gpt-5.5",
           "runtime": {
@@ -1136,7 +1321,7 @@
             "toolPolicy": "read-only",
             "outputMode": "text"
           },
-          "reason": "Top-tier reliability for complex logic flaws in code or architecture.",
+          "reason": "Flagship alternative when DeepSeek Pro is not enough for validated deep exploit chains.",
           "verification": {
             "lastVerifiedAt": "2026-05-29T17:08:03Z",
             "sourceRefs": [
@@ -1149,8 +1334,10 @@
       },
       "fallbackChoiceOrder": [
         "default",
+        "grok_intel",
         "log_analysis",
-        "deep_forensics"
+        "deep_forensics",
+        "gpt_forensics"
       ],
       "requiredCapabilities": {
         "reasoning": "high",
@@ -1162,13 +1349,19 @@
         "outputModalities": [
           "text"
         ]
-      }
+      },
+      "catalogLane": "text"
     },
     "content": {
       "profile": "content",
       "displayName": "Content Generation",
       "description": "Model selection optimized for prose, marketing, and programmatic SEO.",
       "category": "creative",
+      "tags": [
+        "cre",
+        "wrt",
+        "bat"
+      ],
       "aliases": [
         "writing",
         "creative",
@@ -1259,86 +1452,18 @@
         "outputModalities": [
           "text"
         ]
-      }
-    },
-    "summarization": {
-      "profile": "summarization",
-      "displayName": "Summarization",
-      "description": "Optimized for information compression and document triage.",
-      "category": "analysis",
-      "aliases": [
-        "triage",
-        "digest",
-        "tl-dr"
-      ],
-      "defaultChoice": "default",
-      "choices": {
-        "default": {
-          "choice": "default",
-          "displayName": "Fast Digest",
-          "description": "Quickly summarizes standard documents and meeting transcripts.",
-          "provider": "anthropic",
-          "modelId": "claude-haiku-4-5",
-          "runtime": {
-            "instructionTier": "default",
-            "backend": "openrouter",
-            "reasoningEffort": "low",
-            "toolPolicy": "none",
-            "outputMode": "text"
-          },
-          "reason": "Fast and capable at reading comprehension and triage summaries.",
-          "verification": {
-            "lastVerifiedAt": "2026-05-29T17:08:03Z",
-            "sourceRefs": [
-              "vendor:anthropic",
-              "openrouter:catalog"
-            ]
-          },
-          "modelStatus": "verified"
-        },
-        "mega_context": {
-          "choice": "mega_context",
-          "displayName": "Mega Context Summarization",
-          "description": "Compressing entire repositories, books, or multi-year document histories.",
-          "provider": "google",
-          "modelId": "gemini-3.5-flash",
-          "runtime": {
-            "reasoningEffort": "high",
-            "toolPolicy": "none",
-            "outputMode": "text"
-          },
-          "reason": "When input exceeds ~200k tokens, Gemini 3.5 Flash long context handles needle-in-haystack retrieval best.",
-          "verification": {
-            "lastVerifiedAt": "2026-05-29T17:08:03Z",
-            "sourceRefs": [
-              "vendor:google",
-              "openrouter:catalog"
-            ]
-          },
-          "modelStatus": "verified"
-        }
       },
-      "fallbackChoiceOrder": [
-        "default",
-        "mega_context"
-      ],
-      "requiredCapabilities": {
-        "reasoning": "low",
-        "cost": "low",
-        "context": "large",
-        "inputModalities": [
-          "text"
-        ],
-        "outputModalities": [
-          "text"
-        ]
-      }
+      "catalogLane": "text"
     },
     "vision": {
       "profile": "vision",
       "displayName": "Vision & Multimodal",
       "description": "Model selection for image analysis, OCR, and spatial reasoning.",
       "category": "multimodal",
+      "tags": [
+        "vis",
+        "anl"
+      ],
       "aliases": [
         "image",
         "ocr",
@@ -1406,13 +1531,17 @@
         "outputModalities": [
           "text"
         ]
-      }
+      },
+      "catalogLane": "image"
     },
     "local": {
       "profile": "local",
       "displayName": "Local / CPU (no OpenRouter)",
       "description": "On-device inference only — llama.cpp GGUF or Transformers.js on CPU. Does not call OpenRouter or any cloud LLM API.",
       "category": "local",
+      "tags": [
+        "loc"
+      ],
       "aliases": [
         "weak",
         "offline",
@@ -1512,13 +1641,23 @@
         "outputModalities": [
           "text"
         ]
-      }
+      },
+      "catalogLane": "text"
     },
     "coding": {
       "profile": "coding",
       "displayName": "Coding",
       "description": "Model selection optimized for software engineering: implementation, refactors, reviews, and debugging.",
       "category": "engineering",
+      "tags": [
+        "cod",
+        "eng",
+        "agt",
+        "wf",
+        "eco",
+        "ds",
+        "mmx"
+      ],
       "aliases": [
         "dev",
         "developer",
@@ -1542,10 +1681,10 @@
       "choices": {
         "default": {
           "choice": "default",
-          "displayName": "Coding Default",
-          "description": "Recommended model for general coding work.",
-          "provider": "anthropic",
-          "modelId": "claude-sonnet-4-5",
+          "displayName": "Coding Default (DeepSeek V4 Flash)",
+          "description": "First choice for agentic coding loops; escalate to deepseek_pro on failure.",
+          "provider": "deepseek",
+          "modelId": "deepseek-v4-flash",
           "modelStatus": "verified",
           "runtime": {
             "instructionTier": "default",
@@ -1554,53 +1693,78 @@
             "toolPolicy": "read-only",
             "outputMode": "text"
           },
-          "reason": "Strong instruction following and code quality for everyday engineering tasks.",
+          "reason": "Best cost/performance for most coding, patches, and multi-step agent steps.",
           "verification": {
-            "lastVerifiedAt": "2026-05-29T20:57:08Z",
+            "lastVerifiedAt": "2026-06-01T12:00:00Z",
             "sourceRefs": [
-              "vendor:anthropic",
+              "vendor:deepseek",
               "openrouter:catalog"
             ]
           }
         },
-        "openai_coding": {
-          "choice": "openai_coding",
-          "displayName": "OpenAI Coding",
-          "description": "OpenAI alternative for coding workloads.",
-          "provider": "openai",
-          "modelId": "gpt-4.1",
+        "deepseek_pro": {
+          "choice": "deepseek_pro",
+          "displayName": "DeepSeek V4 Pro (escalation)",
+          "description": "Escalation tier when Flash fails or for hard architecture and bug investigation.",
+          "provider": "deepseek",
+          "modelId": "deepseek-v4-pro",
+          "modelStatus": "verified",
+          "runtime": {
+            "instructionTier": "reasoning",
+            "backend": "openrouter",
+            "reasoningEffort": "high",
+            "toolPolicy": "read-only",
+            "outputMode": "text"
+          },
+          "reason": "Use after Flash struggles on complex refactors, debugging, or long-horizon tasks.",
+          "verification": {
+            "lastVerifiedAt": "2026-06-01T12:00:00Z",
+            "sourceRefs": [
+              "vendor:deepseek",
+              "openrouter:catalog"
+            ]
+          }
+        },
+        "claude_coding": {
+          "choice": "claude_coding",
+          "displayName": "Claude Coding",
+          "description": "Premium Claude option for highest instruction-following quality.",
+          "provider": "anthropic",
+          "modelId": "claude-sonnet-4-5",
           "modelStatus": "verified",
           "runtime": {
+            "instructionTier": "default",
+            "backend": "openrouter",
             "reasoningEffort": "high",
             "toolPolicy": "read-only",
             "outputMode": "text"
           },
-          "reason": "Solid alternative for multi-file edits and API-heavy codebases.",
+          "reason": "Strong instruction following and code quality when cost is secondary.",
           "verification": {
             "lastVerifiedAt": "2026-05-29T20:57:08Z",
             "sourceRefs": [
-              "vendor:openai",
+              "vendor:anthropic",
               "openrouter:catalog"
             ]
           }
         },
-        "deepseek_coding": {
-          "choice": "deepseek_coding",
-          "displayName": "Budget Coding",
-          "description": "Lower-cost coding option for simpler changes.",
-          "provider": "deepseek",
-          "modelId": "deepseek-chat",
+        "openai_coding": {
+          "choice": "openai_coding",
+          "displayName": "OpenAI Coding",
+          "description": "OpenAI alternative for coding workloads.",
+          "provider": "openai",
+          "modelId": "gpt-4.1",
           "modelStatus": "verified",
           "runtime": {
-            "reasoningEffort": "medium",
+            "reasoningEffort": "high",
             "toolPolicy": "read-only",
             "outputMode": "text"
           },
-          "reason": "Cost-effective for smaller patches, tests, and boilerplate generation.",
+          "reason": "Solid alternative for multi-file edits and API-heavy codebases.",
           "verification": {
             "lastVerifiedAt": "2026-05-29T20:57:08Z",
             "sourceRefs": [
-              "vendor:deepseek",
+              "vendor:openai",
               "openrouter:catalog"
             ]
           }
@@ -1631,16 +1795,26 @@
       },
       "fallbackChoiceOrder": [
         "default",
+        "deepseek_pro",
+        "claude_coding",
         "openai_coding",
-        "deepseek_coding",
         "minimax_coding"
-      ]
+      ],
+      "catalogLane": "text"
     },
     "extraction": {
       "profile": "extraction",
       "displayName": "Extraction",
       "description": "Model selection for NER, classification, taxonomy mapping, entity extraction, and document labeling.",
       "category": "extraction",
+      "tags": [
+        "xtr",
+        "cls",
+        "str",
+        "json",
+        "bat",
+        "eco"
+      ],
       "aliases": [
         "ner",
         "classify",
@@ -1714,13 +1888,21 @@
       "fallbackChoiceOrder": [
         "default",
         "anthropic_extraction"
-      ]
+      ],
+      "catalogLane": "text"
     },
     "routing": {
       "profile": "routing",
       "displayName": "Routing",
       "description": "Ultra-low-cost models for intent classification, record routing, and workflow branching.",
       "category": "routing",
+      "tags": [
+        "rte",
+        "tri",
+        "cls",
+        "str",
+        "eco"
+      ],
       "aliases": [
         "router",
         "classifier",
@@ -1789,7 +1971,238 @@
       "fallbackChoiceOrder": [
         "default",
         "safer_routing"
-      ]
+      ],
+      "catalogLane": "text"
+    },
+    "vol": {
+      "profile": "vol",
+      "displayName": "Volume Agent",
+      "description": "High-volume economic agent lane: Flash for default agent, coding, extraction, and workflows; Pro for hard reasoning and final-quality answers.",
+      "category": "agent",
+      "tags": [
+        "ds",
+        "eco",
+        "agt",
+        "wf",
+        "cod",
+        "anl",
+        "rsn",
+        "str",
+        "json",
+        "xtr",
+        "cls",
+        "pro",
+        "rev"
+      ],
+      "aliases": [
+        "agent-volume",
+        "flash-tier",
+        "volume-agent",
+        "final-answer"
+      ],
+      "defaultChoice": "default",
+      "choices": {
+        "default": {
+          "choice": "default",
+          "displayName": "DeepSeek Default (V4 Flash)",
+          "description": "Default app, agent, and workflow model with thinking enabled.",
+          "provider": "deepseek",
+          "modelId": "deepseek-v4-flash",
+          "modelStatus": "verified",
+          "runtime": {
+            "instructionTier": "default",
+            "backend": "openrouter",
+            "reasoningEffort": "medium",
+            "toolPolicy": "write-with-approval",
+            "outputMode": "text"
+          },
+          "reason": "Cheap, fast, and strong enough for most coding, extraction, routing, summaries, and agent steps (1M context).",
+          "verification": {
+            "lastVerifiedAt": "2026-06-01T12:00:00Z",
+            "sourceRefs": [
+              "vendor:deepseek",
+              "openrouter:catalog",
+              "vendor:deepseek-api-docs"
+            ]
+          },
+          "secondary": {
+            "kind": "choice",
+            "choice": "pro",
+            "reason": "Escalation to Pro when Flash default is insufficient for agent steps."
+          }
+        },
+        "flash": {
+          "choice": "flash",
+          "displayName": "V4 Flash (non-thinking)",
+          "description": "Classification, extraction, and JSON tasks without expensive reasoning.",
+          "provider": "deepseek",
+          "modelId": "deepseek-v4-flash",
+          "modelStatus": "verified",
+          "runtime": {
+            "instructionTier": "default",
+            "backend": "openrouter",
+            "reasoningEffort": "minimal",
+            "toolPolicy": "none",
+            "outputMode": "schema",
+            "temperature": 0
+          },
+          "reason": "Structured extraction and high-volume tasks that do not need deep reasoning.",
+          "verification": {
+            "lastVerifiedAt": "2026-06-01T12:00:00Z",
+            "sourceRefs": [
+              "vendor:deepseek",
+              "openrouter:catalog"
+            ]
+          }
+        },
+        "pro": {
+          "choice": "pro",
+          "displayName": "V4 Pro (hard reasoning)",
+          "description": "Complex reasoning, architecture, bug investigation, and long-context analysis.",
+          "provider": "deepseek",
+          "modelId": "deepseek-v4-pro",
+          "modelStatus": "verified",
+          "runtime": {
+            "instructionTier": "reasoning",
+            "backend": "openrouter",
+            "reasoningEffort": "high",
+            "toolPolicy": "read-only",
+            "outputMode": "text"
+          },
+          "reason": "Escalation tier for agentic coding loops and hard multi-step reasoning at a fraction of Opus or GPT-5.5 cost.",
+          "verification": {
+            "lastVerifiedAt": "2026-06-01T12:00:00Z",
+            "sourceRefs": [
+              "vendor:deepseek",
+              "openrouter:catalog",
+              "vendor:deepseek-api-docs"
+            ]
+          }
+        },
+        "final": {
+          "choice": "final",
+          "displayName": "V4 Pro (final answer)",
+          "description": "Professional-grade final review and answer nodes.",
+          "provider": "deepseek",
+          "modelId": "deepseek-v4-pro",
+          "modelStatus": "verified",
+          "runtime": {
+            "instructionTier": "reasoning",
+            "backend": "openrouter",
+            "reasoningEffort": "high",
+            "toolPolicy": "none",
+            "outputMode": "text"
+          },
+          "reason": "Best place to spend DeepSeek quality budget on final outputs.",
+          "verification": {
+            "lastVerifiedAt": "2026-06-01T12:00:00Z",
+            "sourceRefs": [
+              "vendor:deepseek",
+              "openrouter:catalog"
+            ]
+          }
+        }
+      },
+      "fallbackChoiceOrder": [
+        "default",
+        "flash",
+        "pro",
+        "final"
+      ],
+      "requiredCapabilities": {
+        "reasoning": "medium",
+        "coding": true,
+        "context": "xlarge",
+        "cost": "low",
+        "supportsTools": true,
+        "inputModalities": [
+          "text"
+        ],
+        "outputModalities": [
+          "text"
+        ]
+      },
+      "catalogLane": "text"
+    },
+    "sum": {
+      "profile": "sum",
+      "displayName": "Summarization",
+      "description": "Optimized for information compression and document triage.",
+      "category": "analysis",
+      "tags": [
+        "sum",
+        "anl",
+        "tri",
+        "wf"
+      ],
+      "aliases": [
+        "triage",
+        "digest",
+        "tl-dr"
+      ],
+      "defaultChoice": "default",
+      "choices": {
+        "default": {
+          "choice": "default",
+          "displayName": "Fast Digest",
+          "description": "Quickly summarizes standard documents and meeting transcripts.",
+          "provider": "anthropic",
+          "modelId": "claude-haiku-4-5",
+          "runtime": {
+            "instructionTier": "default",
+            "backend": "openrouter",
+            "reasoningEffort": "low",
+            "toolPolicy": "none",
+            "outputMode": "text"
+          },
+          "reason": "Fast and capable at reading comprehension and triage summaries.",
+          "verification": {
+            "lastVerifiedAt": "2026-05-29T17:08:03Z",
+            "sourceRefs": [
+              "vendor:anthropic",
+              "openrouter:catalog"
+            ]
+          },
+          "modelStatus": "verified"
+        },
+        "mega_context": {
+          "choice": "mega_context",
+          "displayName": "Mega Context Summarization",
+          "description": "Compressing entire repositories, books, or multi-year document histories.",
+          "provider": "google",
+          "modelId": "gemini-3.5-flash",
+          "runtime": {
+            "reasoningEffort": "high",
+            "toolPolicy": "none",
+            "outputMode": "text"
+          },
+          "reason": "When input exceeds ~200k tokens, Gemini 3.5 Flash long context handles needle-in-haystack retrieval best.",
+          "verification": {
+            "lastVerifiedAt": "2026-05-29T17:08:03Z",
+            "sourceRefs": [
+              "vendor:google",
+              "openrouter:catalog"
+            ]
+          },
+          "modelStatus": "verified"
+        }
+      },
+      "fallbackChoiceOrder": [
+        "default",
+        "mega_context"
+      ],
+      "requiredCapabilities": {
+        "reasoning": "low",
+        "cost": "low",
+        "context": "large",
+        "inputModalities": [
+          "text"
+        ],
+        "outputModalities": [
+          "text"
+        ]
+      },
+      "catalogLane": "text"
     }
   }
 }