@oonacode/win32-x64 0.3.215 → 0.3.218

package/oona.exe

@@ -1,4 +1,4 @@
 [diffend] Oversized file quarantined before diffing.
 name: package/oona.exe
-size: 112337408 bytes
-sha256: e26f69ed725f4f21babb0665415f587b19b44241345da4f6c453b16e57cd54b2
+size: 112930304 bytes
+sha256: b8ce684367826ad4b100b1601227da40037dda6f9b905a187a59677ed024c38b

package/oonacode-release.json

@@ -1,13 +1,13 @@
 {
     "name":  "oonacode",
-    "version":  "0.3.215",
+    "version":  "0.3.218",
     "channel":  "stable",
     "platform":  "win32",
     "arch":  "x64",
     "mode":  "bun",
     "launcherRuntime":  "bun 1.4.1",
-    "commit":  "0cbb4aef",
-    "builtAt":  "2026-10-06T21:31:50.1269361Z",
+    "commit":  "87e61473",
+    "builtAt":  "2026-10-07T02:24:16.1504503Z",
     "engine":  "engine-store",
     "engineVersion":  "0.3.280",
     "siteOrigin":  "https://oonacode.oonak.ai",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@oonacode/win32-x64",
-  "version": "0.3.215",
+  "version": "0.3.218",
   "description": "OonaCode native binaries for Windows x64: oona.exe (CLI + headless host service), the pinned oonaclaude.exe engine sidecar, local web UI assets. Installed automatically as an optional dependency of `oonacode`.",
   "license": "UNLICENSED",
   "homepage": "https://oonacode.oonak.ai",

package/registry.json

@@ -1300,30 +1300,32 @@
           "notes": "Google lists one model per level — gemini-3.7-flash-high, -medium and -low (fetchAvailableModels, 2026-10-03: 1,048,576 context / 65,536 output each); OonaCode shows ONE model and the chat’s effort picker fills the level (upstreamId template; xhigh and max read as high; no effort = high). Cost = Google’s Gemini API list price for gemini-3.7-flash: $0.75 in / $0.075 cached / $3.75 out per 1M through 2026-12-31, DOUBLING to $1.50 / $0.15 / $7.50 from 2027-01-01 (ai.google.dev/gemini-api/docs/pricing, updated 2026-10-01) — move the cost cells on the model page then."
         },
         {
-          "id": "claude-opus-5-5",
-          "displayName": "Claude Opus 5.5 (Antigravity)",
+          "id": "google-claude-opus-5-5",
+          "displayName": "Claude Opus 5.5 (via Google)",
           "upstreamId": "claude-opus-5-5-{effort:high}",
-          "contextWindow": 200000,
+          "contextWindow": 100000,
           "maxOutputTokens": 128000,
           "supportsTools": true,
           "supportsThinking": true,
           "supportsVision": true,
           "inputPricePerM": 4,
           "outputPricePerM": 20,
-          "notes": "Google-hosted Claude through the operator's Antigravity sign-in, the SAME id as Anthropic's row, so the pool folds the two into one entry whose Providers name both, and a new chat goes to the provider with more quota left (owner, 2026-10-04). Google serves it per level — `agy models` (1.2.16, 2026-10-04) lists -low, -medium and -high — and the chat's effort fills the template (xhigh/max read as high). Window 1,000,000 / 128,000 output — read from Google's own fetchAvailableModels on 2026-10-06 (maxTokens 1000000, maxOutputTokens 128000 for every 5.5 level), but the POOL takes prompts only up to 200,000 on this lane (2026-10-06: Google's prompt cache barely applies — 28k of a 417k prompt was read from cache — while Anthropic's caches 94-99 %, so two ~400k-token chats used the account's whole 5-hour Claude window in 3 minutes; a bigger chat stays on, or returns to, Anthropic's lane); thinking is asked as agy 1.3.0 asks it (budget 0 plus the id's level, recorded 2026-10-06). Cost = Anthropic's list price (Google publishes none for its hosted Claude), so a user pays the same on either provider. Anthropic's price: $4 / $0.20 cached / $20 per 1M."
+          "effortLevels": ["low", "medium", "high"],
+          "notes": "PRIVATE (owner, 2026-10-07): in no list — served only when this exact id is asked (/model <id>, the picker’s Custom model… row, oona -m <id>); never the pool entry, /v1/models or the API list (its catalogue row is unlisted). Claude through the operator’s Antigravity sign-in under its OWN id, so the pool never folds it with Anthropic’s lanes. Takes chats up to 100,000 tokens (a bigger prompt, and any PDF, is declined before any upstream call), effort low / medium / high (higher reads as high; the id is a template filled from the chat’s effort). Google lists 1,000,000 / 128,000 for every 5.5 level (fetchAvailableModels, 2026-10-06) but its prompt cache barely applies (28k of a 417k prompt) and it meters by token cost, so the ceiling is the pool’s own. Thinking is asked as agy 1.3.0 asks it (budget 0 plus the id’s level). Cost = Anthropic’s list price ($4 / $0.20 cached / $20 per 1M) — Google publishes none."
         },
         {
-          "id": "claude-sonnet-5-5",
-          "displayName": "Claude Sonnet 5.5 (Antigravity)",
+          "id": "google-claude-sonnet-5-5",
+          "displayName": "Claude Sonnet 5.5 (via Google)",
           "upstreamId": "claude-sonnet-5-5-{effort:high}",
-          "contextWindow": 200000,
+          "contextWindow": 100000,
           "maxOutputTokens": 128000,
           "supportsTools": true,
           "supportsThinking": true,
           "supportsVision": true,
           "inputPricePerM": 2,
           "outputPricePerM": 10,
-          "notes": "Google-hosted Claude through the operator's Antigravity sign-in, the SAME id as Anthropic's row, so the pool folds the two into one entry whose Providers name both, and a new chat goes to the provider with more quota left (owner, 2026-10-04). Google serves it per level — `agy models` (1.2.16, 2026-10-04) lists -low, -medium and -high — and the chat's effort fills the template (xhigh/max read as high). Window 1,000,000 / 128,000 output — read from Google's own fetchAvailableModels on 2026-10-06 (maxTokens 1000000, maxOutputTokens 128000 for every 5.5 level), but the POOL takes prompts only up to 200,000 on this lane (2026-10-06: Google's prompt cache barely applies — 28k of a 417k prompt was read from cache — while Anthropic's caches 94-99 %, so two ~400k-token chats used the account's whole 5-hour Claude window in 3 minutes; a bigger chat stays on, or returns to, Anthropic's lane); thinking is asked as agy 1.3.0 asks it (budget 0 plus the id's level, recorded 2026-10-06). Cost = Anthropic's list price (Google publishes none for its hosted Claude), so a user pays the same on either provider. Anthropic's price: $2 / $0.20 cached / $10 per 1M."
+          "effortLevels": ["low", "medium", "high"],
+          "notes": "PRIVATE (owner, 2026-10-07): in no list — served only when this exact id is asked (/model <id>, the picker’s Custom model… row, oona -m <id>); never the pool entry, /v1/models or the API list (its catalogue row is unlisted). Claude through the operator’s Antigravity sign-in under its OWN id, so the pool never folds it with Anthropic’s lanes. Takes chats up to 100,000 tokens (a bigger prompt, and any PDF, is declined before any upstream call), effort low / medium / high (higher reads as high; the id is a template filled from the chat’s effort). Google lists 1,000,000 / 128,000 for every 5.5 level (fetchAvailableModels, 2026-10-06) but its prompt cache barely applies (28k of a 417k prompt) and it meters by token cost, so the ceiling is the pool’s own. Thinking is asked as agy 1.3.0 asks it (budget 0 plus the id’s level). Cost = Anthropic’s list price ($2 / $0.20 cached / $10 per 1M) — Google publishes none."
         }
       ],
       "defaultModel": "gemini-3.8-flash",