@bubblebrain-ai/bubble 0.0.60 → 0.0.61

package/dist/model-catalog.js

@@ -88,6 +88,9 @@ const ANTHROPIC_CHAT_LEVELS = ["off"];
 // Grok 4.5 exposes low/medium/high effort with no "off" (thinking always on);
 // the composer model declares no effort support at all.
 const GROK_45_LEVELS = ["low", "medium", "high"];
+// Gemini 4 documents only low/medium (default)/high. Verified live on
+// 2026-10-13: thinkingLevel=minimal returns 400, so there is no off/minimal.
+const GEMINI_4_LEVELS = ["low", "medium", "high"];
 const GEMINI_3_LEVELS = ["low", "medium", "high"];
 const GEMINI_3_FLASH_LEVELS = ["minimal", "low", "medium", "high"];
 // Verified against the native API on 2026-09-02: Gemini 3.7/3.8 Flash reject
@@ -149,6 +152,9 @@ export const BUILTIN_MODELS = [
     // list via fetchGeminiModels (GET /v1beta/models, newest five). Gemini 3
     // thinking support varies by release: 3.7/3.8 Flash support off/low/medium/high
     // while older Flash releases expose minimal/low/medium/high.
+    // Gemini 4 Argon: 1,048,576-token window shared by input, thinking and output.
+    // https://ai.google.dev/gemini-api/docs/models/gemini-4-argon
+    { id: "gemini-4-argon", name: "Gemini 4 Argon", providerId: "google", tier: "strong", reasoningLevels: GEMINI_4_LEVELS, defaultReasoningLevel: "medium", contextWindow: 1048576 },
     { id: "gemini-3.8-flash", name: "Gemini 3.8 Flash", providerId: "google", tier: "fast", reasoningLevels: GEMINI_37_38_FLASH_LEVELS, contextWindow: 1048576 },
     { id: "gemini-3.7-flash", name: "Gemini 3.7 Flash", providerId: "google", tier: "fast", reasoningLevels: GEMINI_37_38_FLASH_LEVELS, contextWindow: 1048576 },
     { id: "gemini-3.5-flash", name: "Gemini 3.5 Flash", providerId: "google", tier: "fast", reasoningLevels: GEMINI_3_FLASH_LEVELS, contextWindow: 1048576 },

package/dist/model-pricing.js

@@ -64,6 +64,23 @@ export const MODEL_PRICING_OVERRIDES = [
             outputPerMillion: 7.5,
         },
     })),
+    // Gemini 4 Argon introductory pricing through 2026-11-30; standard rates from
+    // 2026-12-01. Implicit caching only (95% off input), so no cache-write rate.
+    // https://ai.google.dev/gemini-api/docs/pricing
+    {
+        providerId: "google",
+        modelId: "gemini-4-argon",
+        currency: "USD",
+        inputCacheHitPerMillion: 0.1,
+        inputCacheMissPerMillion: 2,
+        outputPerMillion: 10,
+        effectiveUntil: "2026-12-01T00:00:00Z",
+        original: {
+            inputCacheHitPerMillion: 0.2,
+            inputCacheMissPerMillion: 4,
+            outputPerMillion: 20,
+        },
+    },
     {
         providerId: "deepseek",
         modelId: "deepseek-flash",

package/dist/provider-ai-sdk.d.ts

@@ -50,7 +50,8 @@ export declare function fetchGeminiModels(options: {
 }): Promise<GeminiModelDescriptor[]>;
 /**
  * Ranking: one entry per (version, tier) family — GA ids beat dated previews —
- * then newest version first, pro before flash before flash-lite, top N.
+ * then newest version first, flagship (pro/argon) before flash before
+ * flash-lite, top N.
  */
 export declare function selectLatestGeminiModels(entries: GeminiModelListEntry[], limit?: number): GeminiModelDescriptor[];
 /** Mirrors the static catalog's per-family thinking support. */

package/dist/provider-ai-sdk.js

@@ -137,17 +137,23 @@ function buildCallOptions(messages, chatOptions, providerOptions) {
     };
 }
 /**
- * Gemini 3 models take a graded thinking_level; 2.5-era models take a token
+ * Gemini 3+ models take a graded thinking_level; 2.5-era models take a token
  * budget. "off" (budget 0) is only offered in the catalog for models that
  * accept it (2.5 Flash); xhigh/max/ultra clamp to high.
  */
 function buildGoogleThinkingOptions(modelId, level) {
     if (!level)
         return undefined;
+    const version = geminiVersion(modelId);
+    // Gemini 4 accepts only low/medium/high: MINIMAL is a 400 and thinkingBudget
+    // is undocumented. Side calls (titles, compaction) request "off" by default,
+    // so floor those at low rather than sending a value the API rejects.
+    if (version >= 4 && (level === "off" || level === "minimal"))
+        level = "low";
     if (level === "off")
         return { thinkingConfig: { thinkingBudget: 0 } };
     const clamped = level === "xhigh" || level === "max" || level === "ultra" ? "high" : level;
-    if (modelId.includes("gemini-3")) {
+    if (version >= 3) {
         return { thinkingConfig: { thinkingLevel: clamped, includeThoughts: true } };
     }
     const budgets = { minimal: 512, low: 2048, medium: 8192, high: 24576 };
@@ -459,10 +465,13 @@ export async function fetchGeminiModels(options) {
 }
 /**
  * Ranking: one entry per (version, tier) family — GA ids beat dated previews —
- * then newest version first, pro before flash before flash-lite, top N.
+ * then newest version first, flagship (pro/argon) before flash before
+ * flash-lite, top N.
  */
 export function selectLatestGeminiModels(entries, limit = 5) {
-    const TIER_RANK = { pro: 0, flash: 1, "flash-lite": 2 };
+    // Gemini 4 introduced named flagship tiers (gemini-4-argon). Unknown names
+    // stay excluded rather than guessing whether they are text models.
+    const TIER_RANK = { argon: 0, pro: 0, flash: 1, "flash-lite": 2 };
     const candidates = [];
     for (const entry of entries) {
         const id = (entry.name ?? "").replace(/^models\//, "");
@@ -472,21 +481,23 @@ export function selectLatestGeminiModels(entries, limit = 5) {
             continue;
         if (entry.supportedGenerationMethods && !entry.supportedGenerationMethods.includes("generateContent"))
             continue;
-        const match = /^gemini-(\d+(?:\.\d+)?)-(pro|flash)(-lite)?/.exec(id);
+        const match = /^gemini-(\d+(?:\.\d+)?)-(argon|pro|flash)(-lite)?/.exec(id);
         if (!match)
             continue;
+        const tier = `${match[2]}${match[3] ?? ""}`;
         candidates.push({
             id,
             entry,
             version: Number.parseFloat(match[1]),
-            tierRank: TIER_RANK[`${match[2]}${match[3] ?? ""}`] ?? 3,
+            tier,
+            tierRank: TIER_RANK[tier] ?? 3,
             isPreview: id.includes("preview") || id.includes("exp"),
         });
     }
     // One winner per family: GA over preview, then the shortest (least-suffixed) id.
     const families = new Map();
     for (const candidate of candidates) {
-        const key = `${candidate.version}-${candidate.tierRank}`;
+        const key = `${candidate.version}-${candidate.tier}`;
         const current = families.get(key);
         if (!current
             || (current.isPreview && !candidate.isPreview)
@@ -502,13 +513,19 @@ export function selectLatestGeminiModels(entries, limit = 5) {
         name: candidate.entry.displayName || candidate.id,
         contextWindow: Number.isFinite(candidate.entry.inputTokenLimit) ? candidate.entry.inputTokenLimit : undefined,
         reasoningLevels: geminiReasoningLevels(candidate.id),
-        ...(candidate.tierRank === 0 ? { defaultReasoningLevel: "high" } : {}),
+        // Argon defaults to Google's recommended medium via the catalog.
+        ...(candidate.tier === "pro" ? { defaultReasoningLevel: "high" } : {}),
     }));
 }
+function geminiVersion(modelId) {
+    return Number.parseFloat(/^gemini-(\d+(?:\.\d+)?)/.exec(modelId)?.[1] ?? "0");
+}
 /** Mirrors the static catalog's per-family thinking support. */
 export function geminiReasoningLevels(modelId) {
-    const version = Number.parseFloat(/^gemini-(\d+(?:\.\d+)?)/.exec(modelId)?.[1] ?? "0");
+    const version = geminiVersion(modelId);
     const isPro = modelId.includes("-pro");
+    if (version >= 4)
+        return ["low", "medium", "high"];
     // Live probes show 3.7 and 3.8 Flash share this newer ladder. Keep the
     // exception bounded to verified releases instead of guessing for 3.9+.
     if (!isPro && (version === 3.7 || version === 3.8)) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@bubblebrain-ai/bubble",
-  "version": "0.0.60",
+  "version": "0.0.61",
   "description": "A terminal coding agent",
   "type": "module",
   "engines": {