pi-baseten-provider 1.0.7 → 1.0.9

package/README.md

@@ -69,19 +69,20 @@ pi
 
 | Model | Context | Vision | Reasoning | Input $/M | Output $/M |
 |-------|---------|--------|-----------|-----------|------------|
-| DeepSeek V4 Pro | 131K | ❌ | ✅ | $1.74 | $3.48 |
+| Deepseek V4 Flash 0731 | 1.0M | ❌ | ✅ | $0.13 | $0.26 |
+| Deepseek V4 Pro | 262K | ❌ | ✅ | $1.74 | $3.48 |
 | GLM 4.7 | 200K | ❌ | ✅ | $0.60 | $2.20 |
-| GLM 5 | 203K | ❌ | ✅ | $0.95 | $3.15 |
-| GLM 5.1 | 203K | ❌ | ❌ | $1.30 | $4.30 |
-| GLM 5.2 | 524K | ❌ | ✅ | $1.40 | $4.40 |
+| GLM 5.2 | 1.0M | ❌ | ✅ | $1.40 | $4.40 |
 | GLM 5.2 Fast | 524K | ❌ | ✅ | $2.10 | $6.60 |
-| Inkling | 1.0M | ❌ | ✅ | $1.00 | $4.05 |
-| Kimi K2.5 | 262K | ✅ | ✅ | $0.60 | $3.00 |
-| Kimi K2.6 | 262K | ✅ | ✅ | $0.60 | $3.00 |
-| Kimi K2.7 Code | 262K | ❌ | ✅ | $0.95 | $4.00 |
-| Nemotron Super | 203K | ❌ | ✅ | $0.30 | $0.75 |
+| Inkling | 1.0M | ✅ | ✅ | $1.00 | $4.05 |
+| Inkling Small | 1.0M | ✅ | ✅ | $0.50 | $1.20 |
+| Kimi K2.6 | 262K | ✅ | ✅ | $0.95 | $4.00 |
+| Kimi K2.7 Code | 262K | ✅ | ✅ | $0.95 | $4.00 |
+| Kimi K3 | 1.0M | ✅ | ✅ | $3.00 | $15.00 |
+| Mercury 2 | 8K | ❌ | ✅ | Free | Free |
 | Nemotron Ultra | 203K | ❌ | ✅ | $0.60 | $2.40 |
 | OpenAI GPT 120B | 128K | ❌ | ✅ | $0.10 | $0.50 |
+| SID-1 | 33K | ❌ | ❌ | Free | Free |
 
 *Costs are per million tokens. Prices subject to change — check [baseten.co/pricing](https://www.baseten.co/pricing/) for current pricing.*
 

package/custom-models.json

@@ -1,24 +1 @@
-[
-  {
-    "id": "deepseek-ai/DeepSeek-V4-Pro",
-    "name": "DeepSeek V4 Pro",
-    "reasoning": true,
-    "input": [
-      "text"
-    ],
-    "cost": {
-      "input": 1.74,
-      "output": 3.48,
-      "cacheRead": 0.15,
-      "cacheWrite": 0
-    },
-    "contextWindow": 131000,
-    "maxTokens": 131000,
-    "compat": {
-      "supportsDeveloperRole": true,
-      "supportsStore": false,
-      "maxTokensField": "max_completion_tokens",
-      "thinkingFormat": "openai"
-    }
-  }
-]
+[]

package/deprecated-models.json

@@ -0,0 +1 @@
+{}

package/index.ts

@@ -29,6 +29,7 @@ import { getAgentDir, type ExtensionAPI, type ModelRegistry } from "@earendil-wo
 import modelsData from "./models.json" with { type: "json" };
 import customModelsData from "./custom-models.json" with { type: "json" };
 import patchData from "./patch.json" with { type: "json" };
+import deprecatedData from "./deprecated-models.json" with { type: "json" };
 import fs from "fs";
 import path from "path";
 
@@ -48,6 +49,7 @@ interface JsonModel {
   contextWindow: number;
   maxTokens: number;
   thinkingLevelMap?: {
+    off?: string | null;
     minimal?: string | null;
     low?: string | null;
     medium?: string | null;
@@ -59,8 +61,10 @@ interface JsonModel {
     supportsDeveloperRole?: boolean;
     supportsStore?: boolean;
     maxTokensField?: "max_completion_tokens" | "max_tokens";
-    thinkingFormat?: "openai" | "zai" | "qwen" | "qwen-chat-template";
+    thinkingFormat?: "openai" | "zai" | "qwen" | "qwen-chat-template" | "chat-template";
     supportsReasoningEffort?: boolean;
+    requiresReasoningContentOnAssistantMessages?: boolean;
+    chatTemplateKwargs?: Record<string, unknown>;
   };
 }
 
@@ -123,7 +127,10 @@ function applyPatch(model: JsonModel, patch: PatchEntry): JsonModel {
 function buildModels(base: JsonModel[], custom: JsonModel[], patch: PatchData): JsonModel[] {
   const modelMap = new Map<string, JsonModel>();
 
-  for (const model of base) {
+  // Seed with the base list plus grace-period deprecated models so patch.json
+  // entries apply to deprecated models exactly as while the model was live
+  // (withDeprecated keeps live data on id conflicts).
+  for (const model of withDeprecated(base)) {
     modelMap.set(model.id, model);
   }
 
@@ -177,15 +184,18 @@ function transformApiModel(apiModel: any): JsonModel | null {
     cost: {
       input: toPerM(pricing.prompt),
       output: toPerM(pricing.completion),
-      cacheRead: toPerM(pricing.cache_prompt),
+      cacheRead: toPerM(pricing.input_cache_read ?? pricing.cache_prompt),
       cacheWrite: 0,
     },
     contextWindow: apiModel.context_length || 131072,
     maxTokens: apiModel.max_completion_tokens || 131072,
   };
-  if (features.includes("reasoning_effort")) {
-    model.compat = { ...model.compat, supportsReasoningEffort: true };
-  }
+  model.compat = {
+    supportsDeveloperRole: true,
+    supportsStore: false,
+    maxTokensField: "max_completion_tokens",
+    ...(features.includes("reasoning_effort") ? { supportsReasoningEffort: true } : {}),
+  };
   return model;
 }
 
@@ -260,6 +270,35 @@ function mergeWithEmbedded(liveModels: JsonModel[], embeddedModels: JsonModel[])
   return result;
 }
 
+// Grace period for delisted models. When the provider API stops listing a
+// model, update-models.js moves its last-known definition into
+// deprecated-models.json (stamped with deprecatedAt) instead of dropping it.
+// For 14 days the model keeps working here so in-flight sessions and saved
+// model settings do not break; afterwards it is evicted permanently.
+const DEPRECATED_MODEL_TTL_MS = 14 * 24 * 60 * 60 * 1000;
+
+// Grace-period deprecated models with deprecation metadata stripped.
+function activeDeprecatedModels(): JsonModel[] {
+  const now = Date.now();
+  const result: JsonModel[] = [];
+  for (const entry of Object.values(deprecatedData as Record<string, JsonModel & { deprecatedAt?: string }>)) {
+    if (!entry?.id) continue;
+    const removedAt = Date.parse(entry.deprecatedAt ?? "");
+    if (Number.isNaN(removedAt) || now - removedAt > DEPRECATED_MODEL_TTL_MS) continue;
+    const model = { ...entry } as JsonModel & { deprecatedAt?: string };
+    delete model.deprecatedAt;
+    result.push(model);
+  }
+  return result;
+}
+
+// Append grace-period deprecated models the list does not already have (live data wins).
+function withDeprecated(models: JsonModel[]): JsonModel[] {
+  const seen = new Set(models.map((m) => m.id));
+  const extras = activeDeprecatedModels().filter((m) => !seen.has(m.id));
+  return extras.length > 0 ? [...models, ...extras] : models;
+}
+
 function loadStaleModels(embeddedModels: JsonModel[]): JsonModel[] {
   const cached = loadCachedModels();
   if (!cached || cached.length === 0) return embeddedModels;

package/models.json

@@ -1,19 +1,19 @@
 [
   {
-    "id": "deepseek-ai/DeepSeek-V4-Pro",
-    "name": "Deepseek V4 Pro",
+    "id": "deepseek-ai/DeepSeek-V4-Flash-0731",
+    "name": "Deepseek V4 Flash 0731",
     "reasoning": true,
     "input": [
       "text"
     ],
     "cost": {
-      "input": 1.74,
-      "output": 3.48,
-      "cacheRead": 0,
+      "input": 0.13,
+      "output": 0.26,
+      "cacheRead": 0.028,
       "cacheWrite": 0
     },
-    "contextWindow": 262144,
-    "maxTokens": 262144,
+    "contextWindow": 1048576,
+    "maxTokens": 1048576,
     "compat": {
       "supportsDeveloperRole": true,
       "supportsStore": false,
@@ -22,20 +22,20 @@
     }
   },
   {
-    "id": "zai-org/GLM-4.7",
-    "name": "GLM 4.7",
-    "reasoning": false,
+    "id": "deepseek-ai/DeepSeek-V4-Pro",
+    "name": "Deepseek V4 Pro",
+    "reasoning": true,
     "input": [
       "text"
     ],
     "cost": {
-      "input": 0.6,
-      "output": 2.2,
-      "cacheRead": 0,
+      "input": 1.74,
+      "output": 3.48,
+      "cacheRead": 0.145,
       "cacheWrite": 0
     },
-    "contextWindow": 200000,
-    "maxTokens": 200000,
+    "contextWindow": 262144,
+    "maxTokens": 262144,
     "compat": {
       "supportsDeveloperRole": true,
       "supportsStore": false,
@@ -44,20 +44,20 @@
     }
   },
   {
-    "id": "zai-org/GLM-5",
-    "name": "GLM 5",
+    "id": "zai-org/GLM-4.7",
+    "name": "GLM 4.7",
     "reasoning": false,
     "input": [
       "text"
     ],
     "cost": {
-      "input": 0.95,
-      "output": 3.15,
-      "cacheRead": 0,
+      "input": 0.6,
+      "output": 2.2,
+      "cacheRead": 0.12,
       "cacheWrite": 0
     },
-    "contextWindow": 202800,
-    "maxTokens": 202800,
+    "contextWindow": 200000,
+    "maxTokens": 200000,
     "compat": {
       "supportsDeveloperRole": true,
       "supportsStore": false,
@@ -65,27 +65,6 @@
       "thinkingFormat": "openai"
     }
   },
-  {
-    "id": "zai-org/GLM-5.1",
-    "name": "GLM 5.1",
-    "reasoning": false,
-    "input": [
-      "text"
-    ],
-    "cost": {
-      "input": 1.3,
-      "output": 4.3,
-      "cacheRead": 0,
-      "cacheWrite": 0
-    },
-    "contextWindow": 202800,
-    "maxTokens": 202800,
-    "compat": {
-      "supportsDeveloperRole": true,
-      "supportsStore": false,
-      "maxTokensField": "max_completion_tokens"
-    }
-  },
   {
     "id": "zai-org/GLM-5.2",
     "name": "GLM 5.2",
@@ -96,11 +75,11 @@
     "cost": {
       "input": 1.4,
       "output": 4.4,
-      "cacheRead": 0,
+      "cacheRead": 0.14,
       "cacheWrite": 0
     },
-    "contextWindow": 524288,
-    "maxTokens": 524288,
+    "contextWindow": 1048576,
+    "maxTokens": 262144,
     "compat": {
       "supportsDeveloperRole": true,
       "supportsStore": false,
@@ -118,11 +97,11 @@
     "cost": {
       "input": 2.1,
       "output": 6.6,
-      "cacheRead": 0,
+      "cacheRead": 0.21,
       "cacheWrite": 0
     },
     "contextWindow": 524288,
-    "maxTokens": 524288,
+    "maxTokens": 262144,
     "compat": {
       "supportsDeveloperRole": true,
       "supportsStore": false,
@@ -135,12 +114,13 @@
     "name": "Inkling",
     "reasoning": true,
     "input": [
-      "text"
+      "text",
+      "image"
     ],
     "cost": {
       "input": 1,
       "output": 4.05,
-      "cacheRead": 0,
+      "cacheRead": 0.17,
       "cacheWrite": 0
     },
     "contextWindow": 1048576,
@@ -153,21 +133,21 @@
     }
   },
   {
-    "id": "moonshotai/Kimi-K2.5",
-    "name": "Kimi K2.5",
-    "reasoning": false,
+    "id": "thinkingmachines/inkling-small",
+    "name": "Inkling Small",
+    "reasoning": true,
     "input": [
       "text",
       "image"
     ],
     "cost": {
-      "input": 0.6,
-      "output": 3,
-      "cacheRead": 0,
+      "input": 0.5,
+      "output": 1.2,
+      "cacheRead": 0.1,
       "cacheWrite": 0
     },
-    "contextWindow": 262000,
-    "maxTokens": 262000,
+    "contextWindow": 1048576,
+    "maxTokens": 32768,
     "compat": {
       "supportsDeveloperRole": true,
       "supportsStore": false,
@@ -186,7 +166,7 @@
     "cost": {
       "input": 0.95,
       "output": 4,
-      "cacheRead": 0,
+      "cacheRead": 0.16,
       "cacheWrite": 0
     },
     "contextWindow": 262000,
@@ -203,12 +183,13 @@
     "name": "Kimi K2.7 Code",
     "reasoning": true,
     "input": [
-      "text"
+      "text",
+      "image"
     ],
     "cost": {
       "input": 0.95,
       "output": 4,
-      "cacheRead": 0,
+      "cacheRead": 0.16,
       "cacheWrite": 0
     },
     "contextWindow": 262000,
@@ -221,20 +202,43 @@
     }
   },
   {
-    "id": "nvidia/Nemotron-120B-A12B",
-    "name": "Nemotron Super",
+    "id": "moonshotai/Kimi-K3",
+    "name": "Kimi K3",
+    "reasoning": true,
+    "input": [
+      "text",
+      "image"
+    ],
+    "cost": {
+      "input": 3,
+      "output": 15,
+      "cacheRead": 0.3,
+      "cacheWrite": 0
+    },
+    "contextWindow": 1048576,
+    "maxTokens": 262144,
+    "compat": {
+      "supportsDeveloperRole": true,
+      "supportsStore": false,
+      "maxTokensField": "max_completion_tokens",
+      "thinkingFormat": "openai"
+    }
+  },
+  {
+    "id": "inception/mercury-2",
+    "name": "Mercury 2",
     "reasoning": true,
     "input": [
       "text"
     ],
     "cost": {
-      "input": 0.3,
-      "output": 0.75,
+      "input": 0,
+      "output": 0,
       "cacheRead": 0,
       "cacheWrite": 0
     },
-    "contextWindow": 202800,
-    "maxTokens": 202800,
+    "contextWindow": 8192,
+    "maxTokens": 5000,
     "compat": {
       "supportsDeveloperRole": true,
       "supportsStore": false,
@@ -252,7 +256,7 @@
     "cost": {
       "input": 0.6,
       "output": 2.4,
-      "cacheRead": 0,
+      "cacheRead": 0.12,
       "cacheWrite": 0
     },
     "contextWindow": 202800,
@@ -274,7 +278,7 @@
     "cost": {
       "input": 0.1,
       "output": 0.5,
-      "cacheRead": 0,
+      "cacheRead": 0.1,
       "cacheWrite": 0
     },
     "contextWindow": 128072,
@@ -286,5 +290,27 @@
       "thinkingFormat": "openai",
       "supportsReasoningEffort": true
     }
+  },
+  {
+    "id": "sid/sid-1",
+    "name": "SID-1",
+    "reasoning": true,
+    "input": [
+      "text"
+    ],
+    "cost": {
+      "input": 0,
+      "output": 0,
+      "cacheRead": 0,
+      "cacheWrite": 0
+    },
+    "contextWindow": 32768,
+    "maxTokens": 5000,
+    "compat": {
+      "supportsDeveloperRole": true,
+      "supportsStore": false,
+      "maxTokensField": "max_completion_tokens",
+      "thinkingFormat": "openai"
+    }
   }
 ]

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-baseten-provider",
-  "version": "1.0.7",
+  "version": "1.0.9",
   "description": "Baseten provider extension for pi - Access DeepSeek, Kimi, GLM, MiniMax, Nemotron, and GPT-OSS models through the Baseten Model API",
   "author": "monotykamary",
   "homepage": "https://github.com/monotykamary/pi-baseten-provider#readme",
@@ -30,6 +30,7 @@
   "files": [
     "index.ts",
     "models.json",
+    "deprecated-models.json",
     "custom-models.json",
     "patch.json",
     "scripts/update-models.js"

package/patch.json

@@ -1,54 +1,219 @@
 {
-  "moonshotai/Kimi-K2.6": {
-    "reasoning": true,
-    "cost": {
-      "input": 0.6,
-      "output": 3.0
+  "deepseek-ai/DeepSeek-V4-Flash-0731": {
+    "thinkingLevelMap": {
+      "off": "none",
+      "minimal": null,
+      "low": "low",
+      "medium": null,
+      "high": "high",
+      "xhigh": null,
+      "max": "max"
     },
     "compat": {
-      "thinkingFormat": "qwen-chat-template"
+      "thinkingFormat": "chat-template",
+      "supportsReasoningEffort": false,
+      "requiresReasoningContentOnAssistantMessages": true,
+      "chatTemplateKwargs": {
+        "thinking": {
+          "$var": "thinking.enabled"
+        },
+        "reasoning_effort": {
+          "$var": "thinking.effort"
+        }
+      }
     }
   },
-  "moonshotai/Kimi-K2.5": {
-    "reasoning": true,
+  "deepseek-ai/DeepSeek-V4-Pro": {
+    "thinkingLevelMap": {
+      "off": "none",
+      "minimal": "minimal",
+      "low": "low",
+      "medium": "medium",
+      "high": "high",
+      "xhigh": "xhigh",
+      "max": "max"
+    },
     "compat": {
-      "thinkingFormat": "qwen-chat-template"
+      "thinkingFormat": "openai",
+      "supportsReasoningEffort": true,
+      "requiresReasoningContentOnAssistantMessages": true
     }
   },
   "zai-org/GLM-4.7": {
     "reasoning": true,
+    "thinkingLevelMap": {
+      "off": "none",
+      "minimal": null,
+      "low": null,
+      "medium": null,
+      "high": "high",
+      "xhigh": null,
+      "max": null
+    },
     "compat": {
-      "thinkingFormat": "qwen-chat-template"
+      "thinkingFormat": "qwen-chat-template",
+      "supportsReasoningEffort": false,
+      "requiresReasoningContentOnAssistantMessages": true
     }
   },
-  "zai-org/GLM-5": {
-    "reasoning": true,
+  "zai-org/GLM-5.2": {
+    "thinkingLevelMap": {
+      "off": "none",
+      "minimal": null,
+      "low": null,
+      "medium": null,
+      "high": "high",
+      "xhigh": null,
+      "max": "max"
+    },
     "compat": {
-      "thinkingFormat": "qwen-chat-template"
+      "thinkingFormat": "openai",
+      "supportsReasoningEffort": true,
+      "requiresReasoningContentOnAssistantMessages": true
     }
   },
-  "zai-org/GLM-5.2": {
+  "zai-org/GLM-5.2-Fast": {
+    "thinkingLevelMap": {
+      "off": "none",
+      "minimal": null,
+      "low": null,
+      "medium": null,
+      "high": "high",
+      "xhigh": null,
+      "max": "max"
+    },
     "compat": {
-      "thinkingFormat": "qwen-chat-template"
+      "thinkingFormat": "openai",
+      "supportsReasoningEffort": true,
+      "requiresReasoningContentOnAssistantMessages": true
+    }
+  },
+  "thinkingmachines/inkling": {
+    "thinkingLevelMap": {
+      "off": "none",
+      "minimal": "minimal",
+      "low": "low",
+      "medium": "medium",
+      "high": "high",
+      "xhigh": "xhigh",
+      "max": "max"
     },
+    "compat": {
+      "thinkingFormat": "openai",
+      "supportsReasoningEffort": true
+    }
+  },
+  "thinkingmachines/inkling-small": {
     "thinkingLevelMap": {
+      "off": "none",
       "minimal": "minimal",
       "low": "low",
       "medium": "medium",
       "high": "high",
+      "xhigh": "xhigh",
       "max": "max"
+    },
+    "compat": {
+      "thinkingFormat": "openai",
+      "supportsReasoningEffort": true
+    }
+  },
+  "moonshotai/Kimi-K2.6": {
+    "thinkingLevelMap": {
+      "off": "none",
+      "minimal": null,
+      "low": null,
+      "medium": null,
+      "high": "high",
+      "xhigh": null,
+      "max": null
+    },
+    "compat": {
+      "thinkingFormat": "qwen-chat-template",
+      "supportsReasoningEffort": false,
+      "requiresReasoningContentOnAssistantMessages": true
     }
   },
   "moonshotai/Kimi-K2.7-Code": {
     "thinkingLevelMap": {
+      "off": "none",
+      "minimal": null,
+      "low": null,
+      "medium": null,
+      "high": "high",
+      "xhigh": null,
+      "max": null
+    },
+    "compat": {
+      "thinkingFormat": "qwen-chat-template",
+      "supportsReasoningEffort": false,
+      "requiresReasoningContentOnAssistantMessages": true
+    }
+  },
+  "moonshotai/Kimi-K3": {
+    "thinkingLevelMap": {
+      "off": "none",
+      "minimal": null,
+      "low": "low",
+      "medium": null,
+      "high": "high",
+      "xhigh": null,
+      "max": "max"
+    },
+    "compat": {
+      "thinkingFormat": "openai",
+      "supportsReasoningEffort": true,
+      "requiresReasoningContentOnAssistantMessages": true
+    }
+  },
+  "nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B": {
+    "thinkingLevelMap": {
+      "off": "none",
       "minimal": null,
+      "low": null,
+      "medium": null,
+      "high": "high",
+      "xhigh": null,
+      "max": null
+    },
+    "compat": {
+      "thinkingFormat": "qwen-chat-template",
+      "supportsReasoningEffort": false,
+      "requiresReasoningContentOnAssistantMessages": true
+    }
+  },
+  "openai/gpt-oss-120b": {
+    "thinkingLevelMap": {
+      "off": "none",
+      "minimal": "minimal",
+      "low": "low",
+      "medium": "medium",
+      "high": "high",
+      "xhigh": "xhigh",
+      "max": "max"
+    },
+    "compat": {
+      "thinkingFormat": "openai",
+      "supportsReasoningEffort": true
+    }
+  },
+  "inception/mercury-2": {
+    "reasoning": true,
+    "thinkingLevelMap": {
+      "off": "instant",
+      "minimal": "low",
       "low": "low",
       "medium": "medium",
       "high": "high",
-      "xhigh": null
+      "xhigh": null,
+      "max": null
+    },
+    "compat": {
+      "thinkingFormat": "openai",
+      "supportsReasoningEffort": true
     }
   },
-  "nvidia/Nemotron-120B-A12B": {
-    "reasoning": true
+  "sid/sid-1": {
+    "reasoning": false
   }
 }

package/scripts/update-models.js

@@ -96,16 +96,17 @@ function transformApiModel(apiModel, existingModelsMap) {
     }
     // Update features from API
     const features = apiModel.supported_features || [];
-    existing.reasoning = features.includes('reasoning') ?? existing.reasoning;
-    if (features.includes('vision') && !existing.input.includes('image')) {
-      existing.input = ['text', 'image'];
-    }
+    existing.reasoning = features.includes('reasoning');
+    const hasVision = (apiModel.input_modalities || []).includes('image');
+    existing.input = hasVision ? ['text', 'image'] : ['text'];
     // Update pricing from API
     const pricing = apiModel.pricing || {};
     const inputCost = toPerMillion(pricing.prompt);
     const outputCost = toPerMillion(pricing.completion);
-    if (inputCost !== null && inputCost > 0) existing.cost.input = inputCost;
-    if (outputCost !== null && outputCost > 0) existing.cost.output = outputCost;
+    const cacheReadCost = toPerMillion(pricing.input_cache_read ?? pricing.cache_prompt);
+    if (inputCost !== null) existing.cost.input = inputCost;
+    if (outputCost !== null) existing.cost.output = outputCost;
+    if (cacheReadCost !== null) existing.cost.cacheRead = cacheReadCost;
     return existing;
   }
 
@@ -113,7 +114,7 @@ function transformApiModel(apiModel, existingModelsMap) {
   const features = apiModel.supported_features || [];
   const pricing = apiModel.pricing || {};
   const hasReasoning = features.includes('reasoning');
-  const hasVision = features.includes('vision');
+  const hasVision = (apiModel.input_modalities || []).includes('image');
 
   const inputTypes = ['text'];
   if (hasVision) inputTypes.push('image');
@@ -130,7 +131,7 @@ function transformApiModel(apiModel, existingModelsMap) {
     cost: {
       input: inputCost,
       output: outputCost,
-      cacheRead: 0,
+      cacheRead: toPerMillion(pricing.input_cache_read ?? pricing.cache_prompt) || 0,
       cacheWrite: 0,
     },
     contextWindow: apiModel.context_length || 131072,
@@ -276,6 +277,67 @@ function updateReadme(models) {
 
 // ─── Main ────────────────────────────────────────────────────────────────────
 
+// Grace period for delisted models: update-models.js moves models the API no
+// longer lists into deprecated-models.json (stamped with deprecatedAt) instead
+// of dropping them; the runtime appends them back so sessions and saved model
+// settings keep working, and after 14 days they are evicted permanently.
+const DEPRECATED_MODEL_TTL_MS = 14 * 24 * 60 * 60 * 1000;
+
+/**
+ * Reconcile deprecated-models.json against the freshly fetched model list.
+ * - in old models.json but not the API: moved into the deprecated file
+ *   (deprecatedAt = now; preserved on repeat runs so the grace clock is not reset)
+ * - back in the API: resurrected (dropped from the deprecated file)
+ * - deprecatedAt older than 14 days: evicted permanently
+ * Must run BEFORE the new models.json is written; it reads the old file itself.
+ */
+function updateDeprecatedModels(modelsJsonPath, newModels) {
+  const deprecatedPath = path.join(path.dirname(modelsJsonPath), 'deprecated-models.json');
+
+  let oldModels = [];
+  try {
+    const parsed = JSON.parse(fs.readFileSync(modelsJsonPath, 'utf8'));
+    if (Array.isArray(parsed)) oldModels = parsed;
+  } catch { /* first run: no previous models.json */ }
+
+  let deprecated = {};
+  try {
+    const parsed = JSON.parse(fs.readFileSync(deprecatedPath, 'utf8'));
+    if (parsed && typeof parsed === 'object' && !Array.isArray(parsed)) deprecated = parsed;
+  } catch { /* no graveyard yet */ }
+
+  const currentIds = new Set(newModels.map(m => m.id));
+  const now = new Date().toISOString();
+  const added = [];
+  const resurrected = [];
+  const evicted = [];
+
+  for (const old of oldModels) {
+    if (old && old.id && !currentIds.has(old.id) && !deprecated[old.id]) {
+      deprecated[old.id] = { ...old, deprecatedAt: now };
+      added.push(old.id);
+    }
+  }
+
+  for (const [id, entry] of Object.entries(deprecated)) {
+    if (currentIds.has(id)) {
+      delete deprecated[id];
+      resurrected.push(id);
+      continue;
+    }
+    const removedAt = Date.parse(entry && entry.deprecatedAt ? entry.deprecatedAt : '');
+    if (Number.isNaN(removedAt) || Date.now() - removedAt > DEPRECATED_MODEL_TTL_MS) {
+      delete deprecated[id];
+      evicted.push(id);
+    }
+  }
+
+  if (added.length > 0 || resurrected.length > 0 || evicted.length > 0) {
+    fs.writeFileSync(deprecatedPath, JSON.stringify(deprecated, null, 2) + '\n');
+    console.log('Updated deprecated-models.json ' + JSON.stringify({ added, resurrected, evicted }));
+  }
+}
+
 async function main() {
   try {
     const apiModels = await fetchModels();
@@ -299,6 +361,8 @@ async function main() {
     models.sort((a, b) => a.name.localeCompare(b.name));
 
     // Save models.json (pure API output, no patch/custom baked in)
+    // Move delisted models to deprecated-models.json BEFORE models.json is overwritten
+    updateDeprecatedModels(MODELS_JSON_PATH, models);
     saveJson(MODELS_JSON_PATH, models);
 
     // Build full model list for README: base → patch → custom