pi-baseten-provider 1.0.1 → 1.0.3

package/README.md

@@ -73,7 +73,7 @@ pi
 | 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 | 262K | ❌ | ✅ | $1.40 | $4.40 |
+| GLM 5.2 | 203K | ❌ | ✅ | $1.40 | $4.40 |
 | 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 |

package/index.ts

@@ -165,7 +165,10 @@ function transformApiModel(apiModel: any): JsonModel | null {
   const features: string[] = apiModel.supported_features || [];
   const hasVision = (apiModel.input_modalities || []).includes("image");
   const pricing = apiModel.pricing || {};
-  const toPerM = (v: any) => (typeof v === "string" ? parseFloat(v) : (v || 0)) * 1_000_000;
+  // API returns per-token pricing; convert to $/M and round to 6 decimals.
+  // The ×1e6 multiply produces float noise (e.g. 0.7000000000000001); rounding to
+  // 6 decimals normalizes it and preserves sub-cent cache prices like 0.003.
+  const toPerM = (v: any) => Math.round((typeof v === "string" ? parseFloat(v) : (v || 0)) * 1_000_000 * 1e6) / 1e6;
   const model: JsonModel = {
     id: apiModel.id,
     name: apiModel.name || apiModel.id,
@@ -228,9 +231,20 @@ function mergeWithEmbedded(liveModels: JsonModel[], embeddedModels: JsonModel[])
     const embedded = embeddedMap.get(liveModel.id);
     seen.add(liveModel.id);
     if (embedded) {
+      // Self-heal: live API pricing is authoritative field-by-field. Prefer the
+      // live cost when the API reports it (non-zero); fall back to embedded when
+      // the API is silent (0) so curated cacheRead/cacheWrite isn't clobbered and
+      // providers whose /models endpoint exposes no pricing keep their curated
+      // cost. Curation (reasoning/input/compat/name) still wins via ...embedded.
       result.push({
         ...liveModel,
         ...embedded,
+        cost: {
+          input: liveModel.cost.input || embedded.cost.input,
+          output: liveModel.cost.output || embedded.cost.output,
+          cacheRead: liveModel.cost.cacheRead || embedded.cost.cacheRead,
+          cacheWrite: liveModel.cost.cacheWrite || embedded.cost.cacheWrite,
+        },
         contextWindow: liveModel.contextWindow || embedded.contextWindow,
       });
     } else {

package/models.json

@@ -12,8 +12,8 @@
       "cacheRead": 0,
       "cacheWrite": 0
     },
-    "contextWindow": 131000,
-    "maxTokens": 131000,
+    "contextWindow": 262144,
+    "maxTokens": 262144,
     "compat": {
       "supportsDeveloperRole": true,
       "supportsStore": false,
@@ -99,8 +99,8 @@
       "cacheRead": 0,
       "cacheWrite": 0
     },
-    "contextWindow": 262144,
-    "maxTokens": 262144,
+    "contextWindow": 202720,
+    "maxTokens": 202720,
     "compat": {
       "supportsDeveloperRole": true,
       "supportsStore": false,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-baseten-provider",
-  "version": "1.0.1",
+  "version": "1.0.3",
   "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",

package/scripts/update-models.js

@@ -50,7 +50,9 @@ function saveJson(filePath, data) {
 // Convert per-token pricing from API to per-million-tokens
 function toPerMillion(val) {
   if (val === '' || val === null || val === undefined) return null;
-  return Math.round(parseFloat(val) * 1_000_000 * 100) / 100;
+  // Round to 6 decimals of $/M: normalizes float noise from the ×1e6 multiply
+  // and preserves sub-cent cache prices like 0.003 ($/M).
+  return Math.round(parseFloat(val) * 1_000_000 * 1e6) / 1e6;
 }
 
 // ─── API fetch ───────────────────────────────────────────────────────────────