pi-hypercharm-provider 1.1.2 → 1.1.4

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

@@ -15,7 +15,7 @@ _Hyperoptimized coding models — DeepSeek, GLM, Kimi, Qwen, MiniMax, Gemma, GPT
 
 ## Features
 
-- **20+ AI Models** including DeepSeek V4 Flash/Pro, GLM 5/5.1, Kimi K2.5/K2.6, Qwen3.6/3.7, MiniMax M2.7, Gemma 4, GPT-OSS, and Llama
+- **23+ AI Models** including DeepSeek V4 Flash/Pro, GLM 5/5.1, Kimi K2.5/K2.6, Qwen3.6/3.7, MiniMax M2.7, Gemma 4, GPT-OSS, and Llama
 - **DeepSeek Native Thinking** — Uses the `deepseek` thinking format with `reasoning_effort` and 384K max output for DeepSeek V4 models
 - **OpenAI-compatible API** via Charm Hyper's `/v1/chat/completions` endpoint
 - **Cost Tracking** with per-model pricing from the API
@@ -26,24 +26,27 @@ _Hyperoptimized coding models — DeepSeek, GLM, Kimi, Qwen, MiniMax, Gemma, GPT
 
 | Model | Type | Context | Max Tokens | Input Cost | Output Cost |
 |-------|------|---------|------------|------------|-------------|
-| DeepSeek V4 Flash | Text | 1.0M | 384K | $0.14 | $0.28 |
-| DeepSeek V4 Pro | Text | 1.0M | 384K | $1.74 | $3.48 |
-| Gemma 4 26B A4B | Text + Image | 256K | 26K | $0.12 | $0.43 |
+| DeepSeek V4 Flash | Text | 1.0M | 384K | $0.20 | $0.40 |
+| DeepSeek V4 Pro | Text | 1.0M | 384K | $2.40 | $4.80 |
+| Gemma 4 26B A4B | Text + Image | 256K | 26K | $0.12 | $0.41 |
 | GLM 5.1 | Text + Image | 203K | 131K | $1.40 | $4.40 |
-| GLM-5 | Text | 203K | 20K | $0.77 | $2.39 |
-| GLM-5.2 | Text | 1.0M | 131K | $1.40 | $4.40 |
-| GPT-OSS 120B | Text | 131K | 13K | $0.18 | $0.61 |
-| Kimi K2.5 | Text + Image | 262K | 26K | $0.48 | $2.54 |
+| GLM-5 | Text | 203K | 20K | $0.81 | $2.59 |
+| GLM-5.2 | Text | 1.0M | 128K | $1.40 | $4.40 |
+| GPT-OSS 120B | Text | 131K | 13K | $0.17 | $0.56 |
+| Kimi K2.5 | Text + Image | 262K | 26K | $0.52 | $2.72 |
 | Kimi K2.6 | Text + Image | 262K | 262K | $0.95 | $4.00 |
-| Kimi K2.7 Code | Text | 262K | 262K | $0.95 | $4.00 |
+| Kimi K2.7 Code | Text | 256K | 16K | $0.95 | $4.00 |
+| Kimi K3 | Text | 1.0M | 131K | Free | Free |
 | Llama 3.3 70B Instruct | Text | 128K | 13K | $0.61 | $1.04 |
-| Llama 4 Maverick 17B 128E FP8 | Text | 430K | 43K | $0.27 | $0.89 |
-| MiniMax M2.7 | Text | 205K | 20K | $0.82 | $2.64 |
+| Llama 4 Maverick 17B 128E FP8 | Text | 430K | 43K | $0.28 | $0.93 |
+| MiniMax M2.7 | Text | 205K | 20K | $0.42 | $1.56 |
+| MiniMax M3 | Text | 512K | 512K | Free | Free |
 | Qwen3 Coder 480B A35B INT4 | Text | 106K | 11K | $0.60 | $2.08 |
 | Qwen3 Next 80B A3B | Text | 262K | 26K | $0.12 | $1.14 |
 | Qwen3.6-Flash | Text + Image | 1.0M | 64K | $1.00 | $4.00 |
 | Qwen3.6-Max | Text + Image | 256K | 64K | $2.00 | $12.00 |
 | Qwen3.6-Plus | Text + Image | 1.0M | 64K | $2.00 | $6.00 |
+| Qwen3.7-Flash | Text | 1.0M | 64K | Free | Free |
 | Qwen3.7-Max | Text + Image | 1.0M | 64K | $2.50 | $7.50 |
 | Qwen3.7-Plus | Text | 1.0M | 64K | $1.20 | $4.80 |
 *Costs are per million tokens. Prices subject to change — check [hyper.charm.land](https://hyper.charm.land) for current pricing.*

package/deprecated-models.json

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

package/index.ts

@@ -39,6 +39,7 @@ 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";
 
@@ -48,7 +49,7 @@ interface JsonModel {
   id: string;
   name: string;
   reasoning: boolean;
-  input: string[];
+  input: ("text" | "image")[];
   cost: {
     input: number;
     output: number;
@@ -71,7 +72,7 @@ interface JsonModel {
 interface PatchEntry {
   name?: string;
   reasoning?: boolean;
-  input?: string[];
+  input?: ("text" | "image")[];
   cost?: {
     input?: number;
     output?: number;
@@ -124,7 +125,10 @@ 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);
   }
 
@@ -298,6 +302,35 @@ 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

@@ -7,10 +7,10 @@
       "text"
     ],
     "cost": {
-      "input": 0.14,
-      "output": 0.28,
+      "input": 0.2,
+      "output": 0.4,
       "cacheRead": 0.03,
-      "cacheWrite": 0.03
+      "cacheWrite": 0.04
     },
     "contextWindow": 1000000,
     "maxTokens": 384000,
@@ -38,10 +38,10 @@
       "text"
     ],
     "cost": {
-      "input": 1.74,
-      "output": 3.48,
+      "input": 2.4,
+      "output": 4.8,
       "cacheRead": 0.15,
-      "cacheWrite": 0.15
+      "cacheWrite": 0.2
     },
     "contextWindow": 1000000,
     "maxTokens": 384000,
@@ -70,9 +70,9 @@
       "image"
     ],
     "cost": {
-      "input": 0.12,
-      "output": 0.426,
-      "cacheRead": 0.06,
+      "input": 0.116,
+      "output": 0.408,
+      "cacheRead": 0.058,
       "cacheWrite": 0
     },
     "contextWindow": 256000,
@@ -103,9 +103,9 @@
       "text"
     ],
     "cost": {
-      "input": 0.77,
-      "output": 2.388,
-      "cacheRead": 0.385,
+      "input": 0.81,
+      "output": 2.592,
+      "cacheRead": 0.405,
       "cacheWrite": 0
     },
     "contextWindow": 202752,
@@ -122,10 +122,10 @@
       "input": 1.4,
       "output": 4.4,
       "cacheRead": 0.26,
-      "cacheWrite": 0.14
+      "cacheWrite": 0.28
     },
-    "contextWindow": 1048576,
-    "maxTokens": 131072
+    "contextWindow": 1000000,
+    "maxTokens": 128000
   },
   {
     "id": "gpt-oss-120b",
@@ -135,9 +135,9 @@
       "text"
     ],
     "cost": {
-      "input": 0.18,
-      "output": 0.61,
-      "cacheRead": 0.09,
+      "input": 0.168,
+      "output": 0.562,
+      "cacheRead": 0.084,
       "cacheWrite": 0
     },
     "contextWindow": 131072,
@@ -152,9 +152,9 @@
       "image"
     ],
     "cost": {
-      "input": 0.477,
-      "output": 2.535,
-      "cacheRead": 0.2385,
+      "input": 0.518,
+      "output": 2.72,
+      "cacheRead": 0.259,
       "cacheWrite": 0
     },
     "contextWindow": 262144,
@@ -190,8 +190,24 @@
       "cacheRead": 0.16,
       "cacheWrite": 0.19
     },
-    "contextWindow": 262000,
-    "maxTokens": 262000
+    "contextWindow": 256000,
+    "maxTokens": 16000
+  },
+  {
+    "id": "kimi-k3",
+    "name": "Kimi K3",
+    "reasoning": false,
+    "input": [
+      "text"
+    ],
+    "cost": {
+      "input": 0,
+      "output": 0,
+      "cacheRead": 0,
+      "cacheWrite": 0
+    },
+    "contextWindow": 1048576,
+    "maxTokens": 131072
   },
   {
     "id": "llama-3.3-70b-instruct",
@@ -217,9 +233,9 @@
       "text"
     ],
     "cost": {
-      "input": 0.274,
-      "output": 0.894,
-      "cacheRead": 0.137,
+      "input": 0.284,
+      "output": 0.934,
+      "cacheRead": 0.142,
       "cacheWrite": 0
     },
     "contextWindow": 430000,
@@ -233,14 +249,30 @@
       "text"
     ],
     "cost": {
-      "input": 0.82,
-      "output": 2.64,
-      "cacheRead": 0.41,
+      "input": 0.42,
+      "output": 1.56,
+      "cacheRead": 0.21,
       "cacheWrite": 0.06
     },
     "contextWindow": 204800,
     "maxTokens": 20480
   },
+  {
+    "id": "minimax-m3",
+    "name": "MiniMax M3",
+    "reasoning": false,
+    "input": [
+      "text"
+    ],
+    "cost": {
+      "input": 0,
+      "output": 0,
+      "cacheRead": 0,
+      "cacheWrite": 0
+    },
+    "contextWindow": 512000,
+    "maxTokens": 512000
+  },
   {
     "id": "qwen3-coder-480b-a35b-instruct-int4-mixed-ar",
     "name": "Qwen3 Coder 480B A35B INT4",
@@ -324,6 +356,22 @@
     "contextWindow": 1000000,
     "maxTokens": 64000
   },
+  {
+    "id": "qwen3.7-flash",
+    "name": "Qwen3.7-Flash",
+    "reasoning": false,
+    "input": [
+      "text"
+    ],
+    "cost": {
+      "input": 0,
+      "output": 0,
+      "cacheRead": 0,
+      "cacheWrite": 0
+    },
+    "contextWindow": 1000000,
+    "maxTokens": 64000
+  },
   {
     "id": "qwen3.7-max",
     "name": "Qwen3.7-Max",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-hypercharm-provider",
-  "version": "1.1.2",
+  "version": "1.1.4",
   "description": "HyperCharm provider extension for pi - Access DeepSeek, GLM, Kimi, Qwen, MiniMax, Gemma, and GPT-OSS models through the Charm Hyper API",
   "type": "module",
   "main": "index.ts",

package/scripts/update-models.js

@@ -237,6 +237,67 @@ ${tableRows}`;
 
 // ─── 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.HYPERCHARM_API_KEY;
   if (!apiKey) {
@@ -289,6 +350,8 @@ async function main() {
     }
 
     // Update models.json — curated API data
+    // Move delisted models to deprecated-models.json BEFORE models.json is overwritten
+    updateDeprecatedModels(MODELS_JSON_PATH, apiTransformed);
     fs.writeFileSync(MODELS_JSON_PATH, JSON.stringify(apiTransformed, null, 2) + '\n');
     console.log(`✓ Updated models.json (${apiTransformed.length} models)`);