pi-io-provider 1.1.0 → 1.1.2

package/AGENTS.md

@@ -7,6 +7,7 @@ The following files are **idempotent** and regenerated by `scripts/update-models
 | File | Why it's auto-generated |
 |------|------------------------|
 | `models.json` | Built from the provider API. `update-models.js` fetches models, preserves curated data for known IDs, and writes this file. |
+| `deprecated-models.json` | Graveyard for models the API delisted. update-models.js stamps them with deprecatedAt and pi keeps serving them for a 2-week grace period, then evicts them. Never edit by hand. |
 | `README.md` (model table) | The table under `## Available Models` is replaced in-place by `update-models.js` after merging base models → patch → custom models. |
 
 ## Correct Files to Edit

package/README.md

@@ -19,32 +19,34 @@ A [pi](https://github.com/badlogic/pi-mono) extension that adds [IO Intelligence
 | DeepSeek R1 0528 | `deepseek-ai/DeepSeek-R1-0528` | 128K | 128K | ❌ | ✅ | ✅ | $0.57 | $2.28 |
 | Kimi K2 Thinking | `moonshotai/Kimi-K2-Thinking` | 262K | 262K | ❌ | ✅ | ✅ | $0.60 | $2.50 |
 | Llama 3.2 90B Vision Instruct | `meta-llama/Llama-3.2-90B-Vision-Instruct` | 16K | 16K | ✅ | ❌ | ✅ | $0.34 | $0.34 |
-| Llama 4 Maverick 17B 128E Instruct FP8 | `meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8` | 430K | 430K | ✅ | ❌ | ✅ | $0.27 | $0.89 |
-| Kimi K2.5 | `moonshotai/Kimi-K2.5` | 262K | 262K | ✅ | ❌ | ✅ | $0.51 | $2.67 |
-| Kimi K2.6 | `moonshotai/Kimi-K2.6` | 262K | 262K | ✅ | ❌ | ✅ | $0.95 | $3.98 |
-| DeepSeek V3.2 | `deepseek-ai/DeepSeek-V3.2` | 164K | 164K | ❌ | ❌ | ✅ | $0.91 | $1.77 |
-| DeepSeek V4 Flash | `deepseek-ai/DeepSeek-V4-Flash` | 1.0M | 1.0M | ❌ | ❌ | ✅ | $0.13 | $0.26 |
-| DeepSeek V4 Pro | `deepseek-ai/DeepSeek-V4-Pro` | 1.0M | 600K | ❌ | ❌ | ✅ | $1.28 | $2.56 |
-| Gemma 4 26B A4B | `google/gemma-4-26b-a4b-it` | 262K | 262K | ❌ | ❌ | ✅ | $0.12 | $0.43 |
-| Qwen3 Coder 480B A35B Instruct INT4 Mixed AR | `Intel/Qwen3-Coder-480B-A35B-Instruct-int4-mixed-ar` | 106K | 106K | ❌ | ❌ | ✅ | $0.60 | $2.08 |
-| Llama 3.3 70B Instruct | `meta-llama/Llama-3.3-70B-Instruct` | 128K | 128K | ❌ | ❌ | ✅ | $0.55 | $1.07 |
-| MiniMax M2.5 | `MiniMaxAI/MiniMax-M2.5` | 197K | 197K | ❌ | ❌ | ✅ | $0.28 | $1.31 |
-| MiniMax M2.7 | `MiniMaxAI/MiniMax-M2.7` | 205K | 205K | ❌ | ❌ | ✅ | $0.86 | $2.72 |
-| Mistral Nemo Instruct 2407 | `mistralai/Mistral-Nemo-Instruct-2407` | 128K | 128K | ❌ | ❌ | ✅ | $0.06 | $0.10 |
+| Llama 4 Maverick 17B 128E Instruct FP8 | `meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8` | 430K | 430K | ✅ | ❌ | ✅ | $0.28 | $0.93 |
+| Kimi K2.5 | `moonshotai/Kimi-K2.5` | 262K | 262K | ✅ | ❌ | ✅ | $0.53 | $2.79 |
+| Kimi K2.6 | `moonshotai/Kimi-K2.6` | 262K | 262K | ✅ | ❌ | ✅ | $0.84 | $3.76 |
+| DeepSeek V3.2 | `deepseek-ai/DeepSeek-V3.2` | 164K | 164K | ❌ | ❌ | ✅ | $0.90 | $1.76 |
+| DeepSeek V4 Flash | `deepseek-ai/DeepSeek-V4-Flash` | 33K | 33K | ❌ | ❌ | ✅ | $0.20 | $0.37 |
+| DeepSeek V4 Pro | `deepseek-ai/DeepSeek-V4-Pro` | 1.0M | 600K | ❌ | ❌ | ✅ | $1.62 | $3.25 |
+| Gemma 4 26B A4B | `google/gemma-4-26b-a4b-it` | 262K | 262K | ❌ | ❌ | ✅ | $0.12 | $0.39 |
+| Qwen3 Coder 480B A35B Instruct INT4 Mixed AR | `Intel/Qwen3-Coder-480B-A35B-Instruct-int4-mixed-ar` | 106K | 106K | ❌ | ❌ | ✅ | $0.58 | $2.25 |
+| Llama 3.3 70B Instruct | `meta-llama/Llama-3.3-70B-Instruct` | 128K | 128K | ❌ | ❌ | ✅ | $0.51 | $1.04 |
+| MiniMax M2.5 | `MiniMaxAI/MiniMax-M2.5` | 197K | 197K | ❌ | ❌ | ✅ | $0.28 | $1.13 |
+| MiniMax M2.7 | `MiniMaxAI/MiniMax-M2.7` | 205K | 205K | ❌ | ❌ | ✅ | $0.42 | $1.62 |
+| Mistral Nemo Instruct 2407 | `mistralai/Mistral-Nemo-Instruct-2407` | 128K | 128K | ❌ | ❌ | ✅ | $0.07 | $0.12 |
 | Kimi K2 Instruct 0905 | `moonshotai/Kimi-K2-Instruct-0905` | 262K | 262K | ❌ | ❌ | ✅ | $0.57 | $2.30 |
-| Kimi K2.7 Code | `moonshotai/Kimi-K2.7-Code` | 262K | 262K | ❌ | ❌ | ✅ | $1.08 | $4.60 |
-| gpt-oss-120b | `openai/gpt-oss-120b` | 131K | 131K | ❌ | ❌ | ✅ | $0.18 | $0.68 |
-| gpt-oss-20b | `openai/gpt-oss-20b` | 64K | 64K | ❌ | ❌ | ✅ | $0.05 | $0.20 |
+| Kimi K2.7 Code | `moonshotai/Kimi-K2.7-Code` | 262K | 262K | ❌ | ❌ | ✅ | $1.10 | $4.70 |
+| Kimi K3 | `moonshotai/Kimi-K3` | 1.0M | 1.0M | ❌ | ❌ | ✅ | $3.28 | $16.30 |
+| gpt-oss-120b | `openai/gpt-oss-120b` | 131K | 131K | ❌ | ❌ | ✅ | $0.19 | $0.70 |
+| gpt-oss-20b | `openai/gpt-oss-20b` | 64K | 64K | ❌ | ❌ | ✅ | $0.06 | $0.22 |
 | Qwen3 Next 80B A3B Instruct | `Qwen/Qwen3-Next-80B-A3B-Instruct` | 262K | 262K | ❌ | ❌ | ✅ | $0.12 | $1.14 |
-| Qwen3.6 27B | `Qwen/Qwen3.6-27B` | 262K | 262K | ❌ | ❌ | ✅ | $0.37 | $3.03 |
-| Qwen3.6 35B A3B | `Qwen/Qwen3.6-35B-A3B` | 262K | 262K | ❌ | ❌ | ✅ | $0.16 | $1.14 |
+| Qwen3.6 27B | `Qwen/Qwen3.6-27B` | 33K | 33K | ❌ | ❌ | ✅ | $0.40 | $3.09 |
+| Qwen3.6 35B A3B | `Qwen/Qwen3.6-35B-A3B` | 262K | 262K | ❌ | ❌ | ✅ | $0.18 | $1.19 |
+| MiMo-V2.5 | `XiaomiMiMo/MiMo-V2.5` | 262K | 262K | ❌ | ❌ | ✅ | $0.20 | $0.64 |
 | GLM-4.5-Air | `zai-org/GLM-4.5-Air` | 131K | 131K | ❌ | ❌ | ✅ | $0.16 | $0.94 |
-| GLM 4.6 | `zai-org/GLM-4.6` | 131K | 131K | ❌ | ❌ | ✅ | $0.85 | $2.75 |
-| GLM 4.7 | `zai-org/GLM-4.7` | 203K | 203K | ❌ | ❌ | ✅ | $0.86 | $2.22 |
+| GLM 4.6 | `zai-org/GLM-4.6` | 131K | 131K | ❌ | ❌ | ✅ | $0.54 | $2.07 |
+| GLM 4.7 | `zai-org/GLM-4.7` | 203K | 203K | ❌ | ❌ | ✅ | $0.86 | $2.24 |
 | GLM 4.7 Flash | `zai-org/GLM-4.7-Flash` | 200K | 200K | ❌ | ❌ | ✅ | $0.08 | $0.42 |
-| GLM 5 | `zai-org/GLM-5` | 203K | 203K | ❌ | ❌ | ✅ | $0.83 | $2.58 |
-| GLM 5.1 | `zai-org/GLM-5.1` | 203K | 33K | ❌ | ❌ | ✅ | $1.38 | $4.38 |
-| GLM 5.2 | `zai-org/GLM-5.2` | 262K | 131K | ❌ | ❌ | ✅ | $1.46 | $4.74 |
+| GLM 5 | `zai-org/GLM-5` | 203K | 203K | ❌ | ❌ | ✅ | $0.86 | $2.78 |
+| GLM 5.1 | `zai-org/GLM-5.1` | 203K | 33K | ❌ | ❌ | ✅ | $1.29 | $4.22 |
+| GLM 5.2 | `zai-org/GLM-5.2` | 262K | 66K | ❌ | ❌ | ✅ | $1.96 | $6.16 |
 
 *Costs are per million tokens. Cache read/write pricing available on most models.*
 

package/deprecated-models.json

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

package/index.ts

@@ -40,6 +40,7 @@ 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";
 
@@ -49,7 +50,7 @@ interface JsonModel {
   id: string;
   name: string;
   reasoning: boolean;
-  input: string[];
+  input: ("text" | "image")[];
   cost: {
     input: number;
     output: number;
@@ -70,7 +71,7 @@ interface JsonModel {
 interface PatchEntry {
   name?: string;
   reasoning?: boolean;
-  input?: string[];
+  input?: ("text" | "image")[];
   cost?: {
     input?: number;
     output?: number;
@@ -121,7 +122,10 @@ 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);
   }
 
@@ -258,6 +262,35 @@ 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

@@ -1,4 +1,36 @@
 [
+  {
+    "id": "moonshotai/Kimi-K3",
+    "name": "Kimi K3",
+    "reasoning": false,
+    "input": [
+      "text"
+    ],
+    "cost": {
+      "input": 3.28,
+      "output": 16.3,
+      "cacheRead": 1.64,
+      "cacheWrite": 0
+    },
+    "contextWindow": 1048576,
+    "maxTokens": 1048576
+  },
+  {
+    "id": "XiaomiMiMo/MiMo-V2.5",
+    "name": "MiMo-V2.5",
+    "reasoning": false,
+    "input": [
+      "text"
+    ],
+    "cost": {
+      "input": 0.1976,
+      "output": 0.6352,
+      "cacheRead": 0.0988,
+      "cacheWrite": 0
+    },
+    "contextWindow": 262144,
+    "maxTokens": 262144
+  },
   {
     "id": "zai-org/GLM-5.2",
     "name": "GLM 5.2",
@@ -7,13 +39,13 @@
       "text"
     ],
     "cost": {
-      "input": 1.458,
-      "output": 4.74,
-      "cacheRead": 0.729,
+      "input": 1.96,
+      "output": 6.16,
+      "cacheRead": 0.98,
       "cacheWrite": 0
     },
     "contextWindow": 262144,
-    "maxTokens": 131072
+    "maxTokens": 65536
   },
   {
     "id": "moonshotai/Kimi-K2.7-Code",
@@ -23,9 +55,9 @@
       "text"
     ],
     "cost": {
-      "input": 1.078,
-      "output": 4.6,
-      "cacheRead": 0.539,
+      "input": 1.098,
+      "output": 4.7,
+      "cacheRead": 0.549,
       "cacheWrite": 0
     },
     "contextWindow": 262144,
@@ -39,9 +71,9 @@
       "text"
     ],
     "cost": {
-      "input": 0.1628,
-      "output": 1.141313,
-      "cacheRead": 0.0814,
+      "input": 0.1752,
+      "output": 1.19275,
+      "cacheRead": 0.0876,
       "cacheWrite": 0
     },
     "contextWindow": 262140,
@@ -55,13 +87,13 @@
       "text"
     ],
     "cost": {
-      "input": 0.3708,
-      "output": 3.03,
-      "cacheRead": 0.1854,
+      "input": 0.403,
+      "output": 3.09,
+      "cacheRead": 0.2015,
       "cacheWrite": 0
     },
-    "contextWindow": 262140,
-    "maxTokens": 262140
+    "contextWindow": 32768,
+    "maxTokens": 32768
   },
   {
     "id": "MiniMaxAI/MiniMax-M2.7",
@@ -71,9 +103,9 @@
       "text"
     ],
     "cost": {
-      "input": 0.86,
-      "output": 2.72,
-      "cacheRead": 0.43,
+      "input": 0.42,
+      "output": 1.62,
+      "cacheRead": 0.21,
       "cacheWrite": 0
     },
     "contextWindow": 204800,
@@ -87,13 +119,13 @@
       "text"
     ],
     "cost": {
-      "input": 0.13146,
-      "output": 0.26292,
-      "cacheRead": 0.06573,
+      "input": 0.20346,
+      "output": 0.36692,
+      "cacheRead": 0.10173,
       "cacheWrite": 0
     },
-    "contextWindow": 1048576,
-    "maxTokens": 1048576
+    "contextWindow": 32768,
+    "maxTokens": 32768
   },
   {
     "id": "deepseek-ai/DeepSeek-V4-Pro",
@@ -103,9 +135,9 @@
       "text"
     ],
     "cost": {
-      "input": 1.2821,
-      "output": 2.5642,
-      "cacheRead": 0.64105,
+      "input": 1.624,
+      "output": 3.2482,
+      "cacheRead": 0.812,
       "cacheWrite": 0
     },
     "contextWindow": 1048576,
@@ -120,9 +152,9 @@
       "image"
     ],
     "cost": {
-      "input": 0.95,
-      "output": 3.98,
-      "cacheRead": 0.475,
+      "input": 0.84,
+      "output": 3.762,
+      "cacheRead": 0.42,
       "cacheWrite": 0
     },
     "contextWindow": 262142,
@@ -136,9 +168,9 @@
       "text"
     ],
     "cost": {
-      "input": 1.38,
-      "output": 4.38,
-      "cacheRead": 0.69,
+      "input": 1.29,
+      "output": 4.22,
+      "cacheRead": 0.645,
       "cacheWrite": 0
     },
     "contextWindow": 202750,
@@ -152,9 +184,9 @@
       "text"
     ],
     "cost": {
-      "input": 0.285,
-      "output": 1.31,
-      "cacheRead": 0.1425,
+      "input": 0.282,
+      "output": 1.128,
+      "cacheRead": 0.141,
       "cacheWrite": 0
     },
     "contextWindow": 196600,
@@ -169,9 +201,9 @@
       "image"
     ],
     "cost": {
-      "input": 0.51,
-      "output": 2.67,
-      "cacheRead": 0.255,
+      "input": 0.528,
+      "output": 2.79,
+      "cacheRead": 0.264,
       "cacheWrite": 1.1
     },
     "contextWindow": 262144,
@@ -185,9 +217,9 @@
       "text"
     ],
     "cost": {
-      "input": 0.83,
-      "output": 2.58,
-      "cacheRead": 0.415,
+      "input": 0.86,
+      "output": 2.784,
+      "cacheRead": 0.43,
       "cacheWrite": 0
     },
     "contextWindow": 202752,
@@ -201,9 +233,9 @@
       "text"
     ],
     "cost": {
-      "input": 0.9111,
-      "output": 1.7703,
-      "cacheRead": 0.45555,
+      "input": 0.90054,
+      "output": 1.75646,
+      "cacheRead": 0.45027,
       "cacheWrite": 0.5
     },
     "contextWindow": 163840,
@@ -250,7 +282,7 @@
     ],
     "cost": {
       "input": 0.122,
-      "output": 0.428,
+      "output": 0.388,
       "cacheRead": 0.061,
       "cacheWrite": 0
     },
@@ -281,9 +313,9 @@
       "text"
     ],
     "cost": {
-      "input": 0.856,
-      "output": 2.222,
-      "cacheRead": 0.428,
+      "input": 0.864,
+      "output": 2.24,
+      "cacheRead": 0.432,
       "cacheWrite": 0.6
     },
     "contextWindow": 202752,
@@ -330,9 +362,9 @@
       "text"
     ],
     "cost": {
-      "input": 0.178,
-      "output": 0.68,
-      "cacheRead": 0.089,
+      "input": 0.188,
+      "output": 0.7,
+      "cacheRead": 0.094,
       "cacheWrite": 0.04
     },
     "contextWindow": 131072,
@@ -362,9 +394,9 @@
       "text"
     ],
     "cost": {
-      "input": 0.85,
-      "output": 2.75,
-      "cacheRead": 0.3,
+      "input": 0.536,
+      "output": 2.07,
+      "cacheRead": 0.268,
       "cacheWrite": 0.7
     },
     "contextWindow": 131072,
@@ -394,9 +426,9 @@
       "text"
     ],
     "cost": {
-      "input": 0.601,
-      "output": 2.085,
-      "cacheRead": 0.3005,
+      "input": 0.585,
+      "output": 2.245,
+      "cacheRead": 0.2925,
       "cacheWrite": 0.44
     },
     "contextWindow": 106000,
@@ -411,9 +443,9 @@
       "image"
     ],
     "cost": {
-      "input": 0.274,
-      "output": 0.894,
-      "cacheRead": 0.137,
+      "input": 0.284,
+      "output": 0.934,
+      "cacheRead": 0.142,
       "cacheWrite": 0.3
     },
     "contextWindow": 430000,
@@ -427,9 +459,9 @@
       "text"
     ],
     "cost": {
-      "input": 0.0575,
-      "output": 0.0975,
-      "cacheRead": 0.02875,
+      "input": 0.069667,
+      "output": 0.116667,
+      "cacheRead": 0.034834,
       "cacheWrite": 0.04
     },
     "contextWindow": 128000,
@@ -443,9 +475,9 @@
       "text"
     ],
     "cost": {
-      "input": 0.047,
-      "output": 0.198,
-      "cacheRead": 0.0235,
+      "input": 0.059,
+      "output": 0.216,
+      "cacheRead": 0.0295,
       "cacheWrite": 0.03
     },
     "contextWindow": 64000,
@@ -459,9 +491,9 @@
       "text"
     ],
     "cost": {
-      "input": 0.5526,
-      "output": 1.0746,
-      "cacheRead": 0.2763,
+      "input": 0.5126,
+      "output": 1.0446,
+      "cacheRead": 0.2563,
       "cacheWrite": 0.2
     },
     "contextWindow": 128000,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-io-provider",
-  "version": "1.1.0",
+  "version": "1.1.2",
   "description": "IO Intelligence provider extension for pi - Access DeepSeek, Kimi, GLM, Llama, Qwen, Mistral, and more through the IO Intelligence API",
   "type": "module",
   "main": "index.ts",

package/scripts/update-models.js

@@ -328,6 +328,67 @@ MIT
 
 // ─── 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() {
   const apiKey = process.env.IOINTELLIGENCE_API_KEY;
   if (!apiKey) {
@@ -362,6 +423,8 @@ async function main() {
     console.log(`Converted ${models.length} models`);
 
     // 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_PATH, models);
     fs.writeFileSync(MODELS_PATH, JSON.stringify(models, null, 2) + '\n');
     console.log(`✓ Saved ${models.length} models to models.json`);