feat: introduced dynamic version discovery and rate limit parsing for google

- Added dynamic Antigravity version discovery with environment variable overrides and fallback endpoint support.
- Introduced structured Google RPC `ErrorInfo` rate limit reason parsing and backoff classification.
- Added `gemini-3.7-flash` and `deepseek-v4-pro:preview` model configurations alongside updated pricing.
- Removed system instruction injection logic and dropped unsigned thinking blocks for Antigravity requests.

docs/environment-variables.md

@@ -246,7 +246,10 @@ OAuth host chain: `KIMI_CODE_OAUTH_HOST` → `KIMI_OAUTH_HOST` → `https://auth
 | Variable                    | Default / behavior                                              |
 | --------------------------- | --------------------------------------------------------------- |
 | `PI_AI_GEMINI_CLI_VERSION`  | Overrides Gemini CLI user-agent version tag (`0.46.0` if unset) |
-| `PI_AI_ANTIGRAVITY_VERSION` | Overrides Antigravity hub user-agent version (`2.1.4` if unset) |
+| `PI_AI_ANTIGRAVITY_VERSION` | Overrides the auto-discovered Antigravity hub user-agent version; when unset and discovery fails, the fallback is `2.8.0` |
+| `PI_AI_ANTIGRAVITY_CL`      | Overrides Antigravity hub user-agent build changelist (`963137146` if unset) |
+| `PI_AI_ANTIGRAVITY_OS`      | Overrides Antigravity hub user-agent os_type (pinned `darwin` if unset) |
+| `PI_AI_ANTIGRAVITY_ARCH`    | Overrides Antigravity hub user-agent arch (pinned `arm64` if unset) |
 
 ### GitLab Duo
 

docs/provider-quirks.md

@@ -246,8 +246,8 @@ Google Cloud Code Assist (CCA) transport wrapper accessing Gemini and Claude mod
 - **Function Calling Config Mode**: Defaults to `functionCallingConfig: { mode: "VALIDATED" }` for Antigravity in `buildRequest`. Claude models on Antigravity force `VALIDATED` mode even when context contains no declared tools (`isClaudeModel`). Single named tool choice (`options.toolChoice`) sets `mode: "ANY"` with `allowedFunctionNames: [...]`.
 - **Provider Protocol & Request Envelope**:
   - **Endpoints**: `google-gemini-cli` defaults to `https://cloudcode-pa.googleapis.com`. `google-antigravity` uses auto-failover across `https://daily-cloudcode-pa.googleapis.com` (primary) and `https://daily-cloudcode-pa.sandbox.googleapis.com` (sandbox), persisting `lastGoodEndpoint` in `AntigravityProviderSessionState`.
-  - **Headers & User-Agent**: `google-gemini-cli` sends `getGeminiCliHeaders()` (`GeminiCLI/0.46.0/<modelId> (platform; arch; terminal)`). `google-antigravity` sends `getAntigravityUserAgent()` (`antigravity/hub/2.1.4 <os>/<arch>`). Reasoning Claude models on Antigravity send `anthropic-beta: interleaved-thinking-2025-05-14` (`needsClaudeThinkingBetaHeader`).
-  - **System Instructions**: Antigravity tags system instructions with `role: "user"`. Claude and Gemini 3 models prepend `ANTIGRAVITY_SYSTEM_INSTRUCTION` ("You are Antigravity, a powerful agentic AI coding assistant...") via `shouldInjectAntigravitySystemInstruction`.
+  - **Headers & User-Agent**: `google-gemini-cli` sends `getGeminiCliHeaders()` (`GeminiCLI/0.46.0/<modelId> (platform; arch; terminal)`). `google-antigravity` sends `getAntigravityUserAgent()` (`antigravity/hub/<version> (aidev_client; os_type=<os>; arch=<arch>; cl=<cl>)`); the backend gates newer models (e.g. gemini-3.7-flash) on the client version. Reasoning Claude models on Antigravity send `anthropic-beta: interleaved-thinking-2025-05-14` (`needsClaudeThinkingBetaHeader`).
+  - **System Instructions**: Antigravity tags system instructions with `role: "user"`. No identity prompt is injected — the backend accepts arbitrary system instructions on all routes (verified against gemini-3.x and Claude wire ids).
   - **Request Envelope & Session State**: Antigravity wraps requests in `buildAntigravityRequestEnvelope`: `project` (projectId), `requestId` (`agent/<agentId>/<ts>/<trajectoryId>/<step>`), `userAgent` (`antigravity`), `requestType` (`agent`), and `labels` (`last_step_index`, `model_enum`, `trajectory_id`, `used_claude`, `used_claude_conservative`, `last_execution_id`). State maintains monotonic `stepIndex`, persistent `agentId`, `trajectoryId`, and signed-decimal `sessionId` (`deriveAntigravitySessionId`).
   - **Wire Profiles**: `getAntigravityModelWireProfile` (`packages/catalog/src/wire/gemini-headers.ts`) maps wire IDs to `maxOutputTokens` and `model_enum`. Claude wire IDs cap `maxOutputTokens` at `64000` (backend rejects >64000 with 400).
 - **Thinking Configuration & Wire Suppression**: Gemini 2.x models send `thinkingConfig.thinkingBudget`, while Gemini 3 models send `thinkingConfig.thinkingLevel`. When reasoning is disabled for models with `thinking.suppressWhenOff`, `buildRequest` emits explicit wire suppression (`includeThoughts: false` with level/budget). Omitting `thinkingConfig` causes CCA to re-apply server defaults and silently bill thinking tokens.
@@ -892,7 +892,7 @@ The Google Antigravity provider (`google-antigravity`) routes requests to Google
 
 ### Special casings
 - **Validated Function Calling Default**: Default tool selection mode in `buildRequest` (`packages/ai/src/providers/google-gemini-cli.ts`) is `VALIDATED` (`functionCallingConfig: { mode: "VALIDATED" }`). Claude models on Antigravity always force `VALIDATED` tool mode even when no tools are declared (`packages/ai/src/providers/google-gemini-cli.ts`).
-- **System Instruction & Request Envelope**: `shouldInjectAntigravitySystemInstruction` in `packages/ai/src/providers/google-gemini-cli.ts` prepends `ANTIGRAVITY_SYSTEM_INSTRUCTION` with `role: "user"` for Claude and Gemini 3 models. `buildAntigravityRequestEnvelope` injects structured `requestId` (`agent/<id>/<ts>/<trajectoryId>/<step>`), `userAgent: "antigravity"`, `requestType: "agent"`, `sessionId`, and `labels` (`model_enum`, `trajectory_id`, `last_step_index`, `last_execution_id`, `used_claude*`) using `getAntigravityModelWireProfile`.
+- **System Instruction & Request Envelope**: Antigravity tags `systemInstruction` with `role: "user"` and sends the caller's prompts unmodified. `buildAntigravityRequestEnvelope` injects structured `requestId` (`agent/<id>/<ts>/<trajectoryId>/<step>`), `userAgent: "antigravity"`, `requestType: "agent"`, `sessionId`, and `labels` (`model_enum`, `trajectory_id`, `last_step_index`, `last_execution_id`, `used_claude*`) using `getAntigravityModelWireProfile`.
 - **Endpoint Auto-Failover**: Operates across `ANTIGRAVITY_DAILY_ENDPOINT` (`https://daily-cloudcode-pa.googleapis.com`) and `ANTIGRAVITY_SANDBOX_ENDPOINT` (`https://daily-cloudcode-pa.sandbox.googleapis.com`) with state-tracked fallback in `getAntigravityProviderSessionState` (`packages/ai/src/providers/google-gemini-cli.ts`).
 
 ### Auth & usage

packages/ai/CHANGELOG.md

@@ -2,6 +2,16 @@
 
 ## [Unreleased]
 
+### Fixed
+
+- Dropped unsigned thinking blocks from Antigravity Claude requests instead of sending them without a signature, preventing HTTP 400 responses when resuming sessions or switching models.
+- Classified Antigravity HTTP 429 responses from structured `google.rpc.ErrorInfo` reasons (`QUOTA_EXHAUSTED`, `RATE_LIMIT_EXCEEDED`, and `INSUFFICIENT_G1_CREDITS_BALANCE`), using retry delays of five minutes or longer to distinguish rotatable quota windows from transient throttling instead of relying only on message regexes.
+
+### Removed
+
+- Removed the Antigravity identity-prompt injection (`ANTIGRAVITY_SYSTEM_INSTRUCTION` and `shouldInjectAntigravitySystemInstruction`): Cloud Code Assist accepts arbitrary system instructions on gemini-3.x and Claude routes (verified live), and the injected stub never matched the real client's system prompt anyway. User system prompts are now sent unmodified (still tagged `role: "user"`).
+
+
 ## [17.3.0] - 2026-08-13
 
 ### Breaking Changes

packages/ai/src/error/rate-limit.ts

@@ -1,3 +1,5 @@
+import { extractRetryHint } from "@oh-my-pi/pi-utils";
+
 /**
  * Rate limit reason classification and backoff calculation utilities.
  * Ported from opencode-antigravity-auth plugin for consistency.
@@ -5,6 +7,7 @@
 
 export type RateLimitReason =
 	| "QUOTA_EXHAUSTED"
+	| "INSUFFICIENT_G1_CREDITS_BALANCE"
 	| "RATE_LIMIT_EXCEEDED"
 	| "CONCURRENT_LIMIT"
 	| "MODEL_CAPACITY_EXHAUSTED"
@@ -67,6 +70,66 @@ const CN_TRANSIENT_CAP_PATTERN =
 // of rotating through the opaque-429 fallback.
 const CN_THROTTLE_PATTERN = /速率(?:限制|过快)|频率(?:过高|过快)|过于频繁|稍后[重再]试/;
 
+const GOOGLE_RPC_ERROR_INFO_TYPE = "type.googleapis.com/google.rpc.ErrorInfo";
+const LONG_RATE_LIMIT_DELAY_MS = 5 * 60 * 1000;
+
+function asRecord(value: unknown): Record<string, unknown> | undefined {
+	return typeof value === "object" && value !== null && !Array.isArray(value)
+		? (value as Record<string, unknown>)
+		: undefined;
+}
+
+function parseJsonBody(errorMessage: string): Record<string, unknown> | undefined {
+	const start = errorMessage.indexOf("{");
+	const end = errorMessage.lastIndexOf("}");
+	if (start < 0 || end < start) return undefined;
+	try {
+		const parsed: unknown = JSON.parse(errorMessage.slice(start, end + 1));
+		return asRecord(parsed);
+	} catch {
+		return undefined;
+	}
+}
+
+/**
+ * Classify structured Google RESOURCE_EXHAUSTED bodies before consulting text.
+ * Cloud Code Assist prefixes the JSON with its HTTP error label, so accept an
+ * embedded top-level object as well as a raw JSON body.
+ */
+function parseGoogleRpcRateLimitReason(errorMessage: string): RateLimitReason | undefined {
+	const body = parseJsonBody(errorMessage);
+	const error = asRecord(body?.error);
+	if (typeof error?.status !== "string" || error.status.trim().toUpperCase() !== "RESOURCE_EXHAUSTED") {
+		return undefined;
+	}
+	if (!Array.isArray(error.details)) return undefined;
+
+	for (const value of error.details) {
+		const detail = asRecord(value);
+		if (detail?.["@type"] !== GOOGLE_RPC_ERROR_INFO_TYPE || typeof detail.reason !== "string") continue;
+		const reason = detail.reason.trim().toUpperCase();
+		switch (reason) {
+			case "QUOTA_EXHAUSTED":
+				return "QUOTA_EXHAUSTED";
+			case "INSUFFICIENT_G1_CREDITS_BALANCE":
+				// Keep Google's specific credit-balance reason available to logs
+				// and callers while treating it as credential-rotatable below.
+				return "INSUFFICIENT_G1_CREDITS_BALANCE";
+			case "RATE_LIMIT_EXCEEDED": {
+				const retryDelayMs = extractRetryHint(undefined, errorMessage);
+				return retryDelayMs !== undefined && retryDelayMs >= LONG_RATE_LIMIT_DELAY_MS
+					? "QUOTA_EXHAUSTED"
+					: "RATE_LIMIT_EXCEEDED";
+			}
+		}
+	}
+	return undefined;
+}
+
+function isQuotaExhaustedReason(reason: RateLimitReason): boolean {
+	return reason === "QUOTA_EXHAUSTED" || reason === "INSUFFICIENT_G1_CREDITS_BALANCE";
+}
+
 /**
  * Classify a rate-limit error message into a reason category.
  * Priority order: explicit details in a resource-exhausted error > QUOTA
@@ -77,6 +140,8 @@ const CN_THROTTLE_PATTERN = /速率(?:限制|过快)|频率(?:过高|过快)|过
  * Explicit details such as "quota exceeded" retain their normal classification.
  */
 export function parseRateLimitReason(errorMessage: string): RateLimitReason {
+	const structuredReason = parseGoogleRpcRateLimitReason(errorMessage);
+	if (structuredReason !== undefined) return structuredReason;
 	const lowerWithStatus = errorMessage.toLowerCase();
 	const lower = lowerWithStatus.replace(RESOURCE_EXHAUSTED_PATTERN, "");
 	const hasResourceExhaustedStatus = lower !== lowerWithStatus;
@@ -162,6 +227,7 @@ export function parseRateLimitReason(errorMessage: string): RateLimitReason {
  */
 export function calculateRateLimitBackoffMs(reason: RateLimitReason): number {
 	switch (reason) {
+		case "INSUFFICIENT_G1_CREDITS_BALANCE":
 		case "QUOTA_EXHAUSTED":
 			return QUOTA_EXHAUSTED_BACKOFF_MS;
 		case "RATE_LIMIT_EXCEEDED":
@@ -215,6 +281,8 @@ export function isUsageLimitStatus(status: number | undefined): boolean {
  *     credentials.
  */
 export function isUsageLimitOutcome(status: number | undefined, message: string | undefined): boolean {
+	const structuredReason = message ? parseGoogleRpcRateLimitReason(message) : undefined;
+	if (structuredReason !== undefined) return isQuotaExhaustedReason(structuredReason);
 	// Concurrency caps are shed-and-backoff, not credential-rotatable — but only
 	// for quota-worded 429 / other statuses. HTTP 402 is categorically an
 	// account-billing cap, so a 402 whose body happens to mention concurrency is
@@ -235,7 +303,7 @@ export function isUsageLimitOutcome(status: number | undefined, message: string
 	const reason = parseRateLimitReason(message);
 	// For the categorical 402 billing cap a concurrency-worded body is still an
 	// exhausted cap (rotate); for 429 / other only QUOTA_EXHAUSTED rotates.
-	return reason === "QUOTA_EXHAUSTED" || (isBillingCapStatus && reason === "CONCURRENT_LIMIT");
+	return isQuotaExhaustedReason(reason) || (isBillingCapStatus && reason === "CONCURRENT_LIMIT");
 }
 
 /**
@@ -272,6 +340,8 @@ export function isOpaqueStatusBody(message: string): boolean {
  * {@link isUsageLimitOutcome} uses it for the account-rotation decision.
  */
 export function matchesUsageLimitText(errorMessage: string): boolean {
+	const structuredReason = parseGoogleRpcRateLimitReason(errorMessage);
+	if (structuredReason !== undefined) return isQuotaExhaustedReason(structuredReason);
 	return (
 		USAGE_LIMIT_PATTERN.test(errorMessage) ||
 		(CN_QUOTA_EXHAUSTED_PATTERN.test(errorMessage) && !CN_TRANSIENT_CAP_PATTERN.test(errorMessage)) ||

packages/ai/src/providers/google-gemini-cli.ts

@@ -8,7 +8,6 @@ import { scheduler } from "node:timers/promises";
 import { type } from "@oh-my-pi/omptype";
 import { calculateCost } from "@oh-my-pi/pi-catalog/models";
 import {
-	ANTIGRAVITY_SYSTEM_INSTRUCTION,
 	getAntigravityModelWireProfile,
 	getAntigravityUserAgent,
 	getGeminiCliHeaders,
@@ -316,13 +315,6 @@ const ANTIGRAVITY_DAILY_ENDPOINT = "https://daily-cloudcode-pa.googleapis.com";
 const ANTIGRAVITY_SANDBOX_ENDPOINT = "https://daily-cloudcode-pa.sandbox.googleapis.com";
 const ANTIGRAVITY_ENDPOINT_FALLBACKS = [ANTIGRAVITY_DAILY_ENDPOINT, ANTIGRAVITY_SANDBOX_ENDPOINT] as const;
 
-export {
-	ANTIGRAVITY_SYSTEM_INSTRUCTION,
-	getAntigravityUserAgent,
-	getGeminiCliHeaders,
-	getGeminiCliUserAgent,
-} from "@oh-my-pi/pi-catalog/wire/gemini-headers";
-
 // Retry configuration
 const MAX_RETRIES = 3;
 const BASE_DELAY_MS = 1000;
@@ -342,11 +334,6 @@ function needsClaudeThinkingBetaHeader(model: Model<"google-gemini-cli">): boole
 	return model.provider === "google-antigravity" && model.id.startsWith("claude-") && model.reasoning;
 }
 
-function shouldInjectAntigravitySystemInstruction(modelId: string): boolean {
-	const normalized = modelId.toLowerCase();
-	return normalized.includes("claude") || normalized.includes("gemini-3");
-}
-
 const optionalCredentialString = type("unknown").pipe(raw => {
 	const out = type("string")(raw);
 	return out instanceof type.errors ? undefined : out;
@@ -1320,14 +1307,6 @@ export function buildRequest(
 		};
 	}
 
-	if (isAntigravity && shouldInjectAntigravitySystemInstruction(model.id)) {
-		const existingParts = request.systemInstruction?.parts ?? [];
-		request.systemInstruction = {
-			role: "user",
-			parts: [{ text: ANTIGRAVITY_SYSTEM_INSTRUCTION }, ...existingParts],
-		};
-	}
-
 	if (context.tools && context.tools.length > 0) {
 		const convertedTools = convertTools(context.tools, model);
 		request.tools = isAntigravity ? normalizeAntigravityTools(convertedTools) : convertedTools;

packages/ai/src/providers/google-shared.ts

@@ -233,6 +233,8 @@ export function convertMessages<T extends GoogleApiType>(model: Model<T>, contex
 			const parts: Part[] = [];
 			// Check if message is from same provider and model - only then keep thinking blocks
 			const isSameProviderAndModel = msg.provider === model.provider && msg.model === model.id;
+			const dropsUnsignedThinking =
+				model.provider === "google-antigravity" && model.id.toLowerCase().includes("claude");
 
 			for (const block of msg.content) {
 				if (block.type === "text") {
@@ -247,6 +249,7 @@ export function convertMessages<T extends GoogleApiType>(model: Model<T>, contex
 					// Skip empty thinking blocks
 					if (!block.thinking || block.thinking.trim() === "") continue;
 					const thoughtSignature = resolveThoughtSignature(isSameProviderAndModel, block.thinkingSignature);
+					if (dropsUnsignedThinking && !thoughtSignature) continue;
 					if (thoughtSignature) {
 						parts.push({
 							thought: true,

packages/ai/src/registry/oauth/google-antigravity.ts

@@ -2,7 +2,7 @@
  * Antigravity OAuth flow (Gemini 3, Claude, GPT-OSS via Google Cloud)
  * Uses different OAuth credentials than google-gemini-cli for access to additional models.
  */
-import { getAntigravityUserAgent } from "@oh-my-pi/pi-catalog/wire/gemini-headers";
+import { getAntigravityUserAgent, getAntigravityVersion } from "@oh-my-pi/pi-catalog/wire/gemini-headers";
 import * as AIError from "../../error";
 import { oauthFetch, runGoogleOAuthLogin, throwIfLoginCancelled } from "./google-oauth-shared";
 import type { OAuthController, OAuthCredentials } from "./types";
@@ -26,10 +26,12 @@ const SCOPES = [
 const AUTH_URL = "https://accounts.google.com/o/oauth2/v2/auth";
 const TOKEN_URL = "https://oauth2.googleapis.com/token";
 const CLOUD_CODE_ENDPOINT = "https://cloudcode-pa.googleapis.com";
-const TIER_LEGACY = "legacy-tier";
+const DAILY_CLOUD_CODE_ENDPOINT = "https://daily-cloudcode-pa.googleapis.com";
+const NODE_API_CLIENT_USER_AGENT = "google-api-nodejs-client/10.3.0";
+const GOOG_API_CLIENT_HEADER = "gl-node/22.21.1";
+const TIER_FREE = "free-tier";
 const PROJECT_ONBOARD_MAX_ATTEMPTS = 5;
 const PROJECT_ONBOARD_INTERVAL_MS = 2000;
-
 interface LoadCodeAssistPayload {
 	cloudaicompanionProject?: string | { id?: string };
 	currentTier?: { id?: string };
@@ -45,35 +47,65 @@ interface LongRunningOperationResponse {
 
 export const ANTIGRAVITY_LOAD_CODE_ASSIST_METADATA = Object.freeze({
 	ideType: "ANTIGRAVITY",
-	platform: "PLATFORM_UNSPECIFIED",
-	pluginType: "GEMINI",
 });
 
-function readProjectId(value: string | { id?: string } | undefined): string | undefined {
+export interface AntigravityOnboardMetadata {
+	ide_type: string;
+	ide_version: string;
+	ide_name: string;
+}
+
+export function getAntigravityOnboardMetadata(): AntigravityOnboardMetadata {
+	return {
+		ide_type: "ANTIGRAVITY",
+		ide_version: getAntigravityVersion(),
+		ide_name: "antigravity",
+	};
+}
+
+function readProjectId(value: unknown): string | undefined {
 	if (typeof value === "string" && value.length > 0) {
-		return value;
+		return value.trim();
 	}
-	if (value && typeof value === "object" && typeof value.id === "string" && value.id.length > 0) {
-		return value.id;
+	if (value && typeof value === "object" && "id" in value && typeof (value as { id?: unknown }).id === "string") {
+		const id = (value as { id: string }).id.trim();
+		if (id.length > 0) return id;
 	}
 	return undefined;
 }
 
-function getDefaultTierId(allowedTiers?: Array<{ id?: string; isDefault?: boolean }>): string {
-	if (!allowedTiers || allowedTiers.length === 0) {
-		return TIER_LEGACY;
+function extractProjectId(payload: unknown): string | undefined {
+	if (!payload || typeof payload !== "object") return undefined;
+	const record = payload as Record<string, unknown>;
+	for (const key of ["cloudaicompanionProject", "projectId", "project"]) {
+		const id = readProjectId(record[key]);
+		if (id) return id;
 	}
-	const defaultTier = allowedTiers.find(tier => tier.isDefault && typeof tier.id === "string" && tier.id.length > 0);
-	if (defaultTier?.id) {
-		return defaultTier.id;
+	return undefined;
+}
+
+function getDefaultTierId(
+	allowedTiers?: Array<{ id?: string; isDefault?: boolean }>,
+	currentTier?: { id?: string },
+): string {
+	if (allowedTiers && allowedTiers.length > 0) {
+		const defaultTier = allowedTiers.find(
+			tier => tier.isDefault && typeof tier.id === "string" && tier.id.trim().length > 0,
+		);
+		if (defaultTier?.id) {
+			return defaultTier.id.trim();
+		}
 	}
-	return TIER_LEGACY;
+	if (currentTier && typeof currentTier.id === "string" && currentTier.id.trim().length > 0) {
+		return currentTier.id.trim();
+	}
+	return TIER_FREE;
 }
 
 async function onboardProjectWithRetries(
 	endpoint: string,
 	headers: Record<string, string>,
-	onboardBody: { tierId: string; metadata: typeof ANTIGRAVITY_LOAD_CODE_ASSIST_METADATA },
+	onboardBody: { tier_id: string; metadata: AntigravityOnboardMetadata },
 	signal: AbortSignal | undefined,
 	onProgress?: (message: string) => void,
 ): Promise<string> {
@@ -104,7 +136,7 @@ async function onboardProjectWithRetries(
 			continue;
 		}
 
-		const projectId = readProjectId(operation.response?.cloudaicompanionProject);
+		const projectId = extractProjectId(operation.response);
 		if (projectId) {
 			return projectId;
 		}
@@ -128,42 +160,65 @@ async function discoverProject(
 	};
 
 	onProgress?.("Checking for existing project...");
-	const endpoint = CLOUD_CODE_ENDPOINT;
 	try {
-		throwIfLoginCancelled(signal);
-		const loadResponse = await oauthFetch(
-			`${endpoint}/v1internal:loadCodeAssist`,
-			{
-				method: "POST",
-				headers,
-				body: JSON.stringify({
-					metadata: ANTIGRAVITY_LOAD_CODE_ASSIST_METADATA,
-				}),
-			},
-			{ provider: "google-antigravity", signal },
-		);
+		let lastErrorText: string | undefined;
+		let lastStatus: number | undefined;
+		let fallbackTierId = TIER_FREE;
+		let loadedSuccessfully = false;
 
-		if (!loadResponse.ok) {
-			const errorText = await loadResponse.text();
-			throw new AIError.OAuthError(
-				`loadCodeAssist failed: ${loadResponse.status} ${loadResponse.statusText}: ${errorText}`,
-				{ kind: "discovery", status: loadResponse.status },
+		for (const endpoint of [DAILY_CLOUD_CODE_ENDPOINT, CLOUD_CODE_ENDPOINT]) {
+			throwIfLoginCancelled(signal);
+			const loadResponse = await oauthFetch(
+				`${endpoint}/v1internal:loadCodeAssist`,
+				{
+					method: "POST",
+					headers,
+					body: JSON.stringify({
+						metadata: ANTIGRAVITY_LOAD_CODE_ASSIST_METADATA,
+					}),
+				},
+				{ provider: "google-antigravity", signal },
 			);
+
+			if (!loadResponse.ok) {
+				lastStatus = loadResponse.status;
+				lastErrorText = await loadResponse.text();
+				continue;
+			}
+
+			loadedSuccessfully = true;
+			const loadPayload = (await loadResponse.json()) as LoadCodeAssistPayload;
+			const existingProject = extractProjectId(loadPayload);
+			if (existingProject) {
+				return existingProject;
+			}
+			fallbackTierId = getDefaultTierId(loadPayload.allowedTiers, loadPayload.currentTier);
 		}
 
-		const loadPayload = (await loadResponse.json()) as LoadCodeAssistPayload;
-		const existingProject = readProjectId(loadPayload.cloudaicompanionProject);
-		if (existingProject) {
-			return existingProject;
+		if (!loadedSuccessfully && lastStatus !== undefined) {
+			throw new AIError.OAuthError(`loadCodeAssist failed: ${lastStatus}: ${lastErrorText || "unknown error"}`, {
+				kind: "discovery",
+				status: lastStatus,
+			});
 		}
 
-		const tierId = getDefaultTierId(loadPayload.allowedTiers);
 		onProgress?.("Provisioning project...");
 		const onboardBody = {
-			tierId,
-			metadata: ANTIGRAVITY_LOAD_CODE_ASSIST_METADATA,
+			tier_id: fallbackTierId,
+			metadata: getAntigravityOnboardMetadata(),
 		};
-		const provisionedProject = await onboardProjectWithRetries(endpoint, headers, onboardBody, signal, onProgress);
+		const onboardHeaders: Record<string, string> = {
+			...headers,
+			"User-Agent": `${headers["User-Agent"]} ${NODE_API_CLIENT_USER_AGENT}`,
+			"X-Goog-Api-Client": GOOG_API_CLIENT_HEADER,
+		};
+		const provisionedProject = await onboardProjectWithRetries(
+			DAILY_CLOUD_CODE_ENDPOINT,
+			onboardHeaders,
+			onboardBody,
+			signal,
+			onProgress,
+		);
 		return provisionedProject;
 	} catch (error) {
 		if (error instanceof AIError.LoginCancelledError || error instanceof AIError.OAuthError) {

packages/ai/src/usage/gemini.ts

@@ -1,4 +1,4 @@
-import { getGeminiCliHeaders } from "../providers/google-gemini-cli";
+import { getGeminiCliHeaders } from "@oh-my-pi/pi-catalog/wire/gemini-headers";
 import type {
 	UsageAmount,
 	UsageFetchContext,

packages/ai/test/google-gemini-cli-alignment.test.ts

@@ -1,7 +1,6 @@
 import { describe, expect, it } from "bun:test";
 import * as geminiCliProvider from "@oh-my-pi/pi-ai/providers/google-gemini-cli";
 import {
-	ANTIGRAVITY_SYSTEM_INSTRUCTION,
 	buildRequest,
 	parseGeminiCliCredentials,
 	shouldRefreshGeminiCliCredentials,
@@ -189,9 +188,70 @@ describe("Google Gemini CLI alignment", () => {
 		expect(payload.request.contents).toEqual([{ role: "user", parts: [{ text: "implement token refresh" }] }]);
 	});
 
+	it("drops only unsigned thinking when replaying Antigravity Claude history", () => {
+		const signedThinking = "signed reasoning";
+		const unsignedThinking = "unsigned reasoning";
+		const signature = "c2lnbmVk";
+		const createThinkingContext = (model: Model<"google-gemini-cli">): Context => ({
+			messages: [
+				{
+					role: "assistant",
+					content: [
+						{ type: "thinking", thinking: signedThinking, thinkingSignature: signature },
+						{ type: "thinking", thinking: unsignedThinking },
+					],
+					api: model.api,
+					provider: model.provider,
+					model: model.id,
+					usage: {
+						input: 0,
+						output: 0,
+						cacheRead: 0,
+						cacheWrite: 0,
+						totalTokens: 0,
+						cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
+					},
+					stopReason: "stop",
+					timestamp: 1,
+				},
+				{ role: "user", content: "continue", timestamp: 2 },
+			],
+		});
+		const claudeModel = buildModel({
+			...createModel("google-antigravity"),
+			id: "claude-sonnet-4-6",
+			name: "Claude Sonnet 4.6",
+			reasoning: true,
+		} as ModelSpec<"google-gemini-cli">);
+		const claudePayload = buildRequest(claudeModel, createThinkingContext(claudeModel), "proj-123", {}, true) as {
+			request: {
+				contents: Array<{
+					role: string;
+					parts: Array<{ text?: string; thought?: boolean; thoughtSignature?: string }>;
+				}>;
+			};
+		};
+		const claudeParts = claudePayload.request.contents.find(content => content.role === "model")?.parts;
+		expect(claudeParts).toEqual([{ thought: true, text: signedThinking, thoughtSignature: signature }]);
+
+		const geminiModel = createModel("google-antigravity");
+		const geminiPayload = buildRequest(geminiModel, createThinkingContext(geminiModel), "proj-123", {}, true) as {
+			request: {
+				contents: Array<{
+					role: string;
+					parts: Array<{ text?: string; thought?: boolean; thoughtSignature?: string }>;
+				}>;
+			};
+		};
+		const geminiParts = geminiPayload.request.contents.find(content => content.role === "model")?.parts ?? [];
+		expect(geminiParts).toContainEqual({ thought: true, text: signedThinking, thoughtSignature: signature });
+		expect(geminiParts.some(part => part.text?.includes(unsignedThinking))).toBe(true);
+	});
+
 	it("keeps antigravity metadata in antigravity request payloads", () => {
 		const model = createModel("google-antigravity");
-		const payload = buildRequest(model, createContext(), "proj-123", {}, true) as {
+		const context: Context = { ...createContext(), systemPrompt: ["be terse"] };
+		const payload = buildRequest(model, context, "proj-123", {}, true) as {
 			request: {
 				sessionId?: string;
 				labels?: Record<string, string>;
@@ -324,30 +384,6 @@ describe("Google Gemini CLI alignment", () => {
 		expect(parameters).toBeDefined();
 		expect(JSON.stringify(parameters)).not.toContain('"patternProperties"');
 	});
-	it("injects ANTIGRAVITY_SYSTEM_INSTRUCTION for gemini-3.1-pro-high and gemini-3.1-pro-low", () => {
-		// Regression test for #1274: shouldInjectAntigravitySystemInstruction checked
-		// "gemini-3-pro-high" (hyphen) but the deployed model IDs use "gemini-3.1-pro-high" (dot),
-		// so the injection was silently skipped and the Cloud Code Assist API returned HTTP 400.
-		for (const modelId of ["gemini-3.1-pro-high", "gemini-3.1-pro-low"] as const) {
-			const model: Model<"google-gemini-cli"> = buildModel({
-				...createModel("google-antigravity"),
-				id: modelId,
-			} as ModelSpec<"google-gemini-cli">);
-			const context: Context = {
-				systemPrompt: ["my instructions"],
-				messages: [{ role: "user", content: "hi", timestamp: Date.now() }],
-			};
-			const payload = buildRequest(model, context, "proj-123", {}, true) as {
-				request: { systemInstruction?: { role?: string; parts: Array<{ text: string }> } };
-			};
-
-			const parts = payload.request.systemInstruction?.parts ?? [];
-			// The antigravity identity header must be injected as the first part.
-			expect(parts[0]?.text).toBe(ANTIGRAVITY_SYSTEM_INSTRUCTION);
-			// The user-supplied system prompt must appear after the single injected part.
-			expect(parts.slice(1).some(p => p.text === "my instructions")).toBe(true);
-		}
-	});
 	it("adds anthropic-beta for Antigravity Claude reasoning models without relying on id suffix", async () => {
 		let requestHeaders: Headers | undefined;
 		const fetchMock: FetchImpl = async (_url, init) => {

packages/ai/test/rate-limit-utils.test.ts

@@ -10,6 +10,30 @@ import {
 	parseRateLimitReason,
 } from "@oh-my-pi/pi-ai/error/rate-limit";
 
+function googleRpc429(reason: string, retryDelay?: string, message = "Resource exhausted"): string {
+	const details: Array<Record<string, string>> = [
+		{
+			"@type": "type.googleapis.com/google.rpc.ErrorInfo",
+			reason,
+			domain: "cloudcode-pa.googleapis.com",
+		},
+	];
+	if (retryDelay) {
+		details.push({
+			"@type": "type.googleapis.com/google.rpc.RetryInfo",
+			retryDelay,
+		});
+	}
+	return `Cloud Code Assist API error (429): ${JSON.stringify({
+		error: {
+			code: 429,
+			message,
+			status: "RESOURCE_EXHAUSTED",
+			details,
+		},
+	})}`;
+}
+
 describe("parseRateLimitReason", () => {
 	it("classifies Google Quota exceeded as QUOTA_EXHAUSTED", () => {
 		expect(
@@ -170,6 +194,54 @@ describe("parseRateLimitReason", () => {
 			),
 		).toBe("QUOTA_EXHAUSTED");
 	});
+
+	it("uses structured QUOTA_EXHAUSTED before capacity message heuristics", () => {
+		const body = googleRpc429(
+			"QUOTA_EXHAUSTED",
+			"21600s",
+			"The model has no capacity available; retry another request later.",
+		);
+		expect(parseRateLimitReason(body)).toBe("QUOTA_EXHAUSTED");
+		expect(isUsageLimitOutcome(429, body)).toBe(true);
+	});
+
+	it("keeps structured RATE_LIMIT_EXCEEDED with a 30s delay transient", () => {
+		const body = googleRpc429("RATE_LIMIT_EXCEEDED", "30s", "Too many requests");
+		expect(parseRateLimitReason(body)).toBe("RATE_LIMIT_EXCEEDED");
+		expect(isUsageLimitOutcome(429, body)).toBe(false);
+		expect(isUsageLimit(Object.assign(new Error(body), { status: 429 }))).toBe(false);
+	});
+
+	it("keeps structured RATE_LIMIT_EXCEEDED without a retry delay transient", () => {
+		const body = googleRpc429("RATE_LIMIT_EXCEEDED", undefined, "Too many requests");
+		expect(parseRateLimitReason(body)).toBe("RATE_LIMIT_EXCEEDED");
+		expect(isUsageLimitOutcome(429, body)).toBe(false);
+	});
+
+	it("treats structured RATE_LIMIT_EXCEEDED with a 6h delay as usage exhaustion", () => {
+		const body = googleRpc429("RATE_LIMIT_EXCEEDED", "21600s", "Too many requests");
+		expect(parseRateLimitReason(body)).toBe("QUOTA_EXHAUSTED");
+		expect(isUsageLimitOutcome(429, body)).toBe(true);
+	});
+
+	it("treats the five-minute structured rate-limit threshold as usage exhaustion", () => {
+		const body = googleRpc429("RATE_LIMIT_EXCEEDED", "300s", "Too many requests");
+		expect(parseRateLimitReason(body)).toBe("QUOTA_EXHAUSTED");
+		expect(isUsageLimitOutcome(429, body)).toBe(true);
+	});
+
+	it("preserves structured INSUFFICIENT_G1_CREDITS_BALANCE while rotating credentials", () => {
+		const body = googleRpc429("INSUFFICIENT_G1_CREDITS_BALANCE", undefined, "Credit balance is unavailable");
+		expect(parseRateLimitReason(body)).toBe("INSUFFICIENT_G1_CREDITS_BALANCE");
+		expect(isUsageLimitOutcome(429, body)).toBe(true);
+		expect(isUsageLimit(Object.assign(new Error(body), { status: 429 }))).toBe(true);
+	});
+
+	it("falls back to existing text heuristics for non-JSON 429 bodies", () => {
+		const body = "Cloud Code Assist API error (429): Too many requests";
+		expect(parseRateLimitReason(body)).toBe("RATE_LIMIT_EXCEEDED");
+		expect(isUsageLimitOutcome(429, body)).toBe(false);
+	});
 });
 
 describe("isUsageLimit", () => {

packages/catalog/CHANGELOG.md

@@ -2,6 +2,20 @@
 
 ## [Unreleased]
 
+### Added
+
+- Added support for the `deepseek-v4-pro:preview` model
+- Added support for the `gemini-3.7-flash` model
+- Added dynamic Antigravity client-version discovery from the official update manifest (darwin/arm64 channel), so version-gated models appear without a code change; `PI_AI_ANTIGRAVITY_VERSION` remains available as an override.
+
+### Removed
+
+- Removed `ANTIGRAVITY_SYSTEM_INSTRUCTION` from `wire/gemini-headers`; the Antigravity transport and web search no longer inject a fake identity prompt.
+
+### Fixed
+
+- Fixed Antigravity discovery missing Gemini 3.7 Flash: Cloud Code Assist gates newer models on the client version in the `User-Agent`, and the pinned `antigravity/hub/2.1.4` was too old. The user-agent now matches the captured 2.8.0 client format (`antigravity/hub/2.8.0 (aidev_client; os_type=darwin; arch=arm64; cl=963137146)`); os_type/arch stay pinned to the darwin/arm64 reference client. Overridable via `PI_AI_ANTIGRAVITY_VERSION` / `PI_AI_ANTIGRAVITY_CL` / `PI_AI_ANTIGRAVITY_OS` / `PI_AI_ANTIGRAVITY_ARCH`.
+
 ## [17.3.1] - 2026-08-13
 
 ### Added

packages/catalog/src/discovery/antigravity.ts

@@ -6,7 +6,7 @@ import {
 	collapseEffortVariants,
 	type VariantCollapseTable,
 } from "../variant-collapse";
-import { getAntigravityUserAgent } from "../wire/gemini-headers";
+import { ensureAntigravityVersion, getAntigravityUserAgent } from "../wire/gemini-headers";
 
 export const ANTIGRAVITY_PRIMARY_ENDPOINT = "https://daily-cloudcode-pa.googleapis.com";
 export const ANTIGRAVITY_SANDBOX_ENDPOINT = "https://daily-cloudcode-pa.sandbox.googleapis.com";
@@ -161,6 +161,10 @@ export interface FetchAntigravityDiscoveryModelsOptions {
 export async function fetchAntigravityDiscoveryModels(
 	options: FetchAntigravityDiscoveryModelsOptions,
 ): Promise<ModelSpec<"google-gemini-cli">[] | null> {
+	if (options.userAgent === undefined) {
+		await ensureAntigravityVersion(fetch, options.signal);
+	}
+
 	const fetcher = discoveryFetch(options.fetcher);
 	const endpoints = options.endpoint
 		? [trimTrailingSlashes(options.endpoint)]

packages/catalog/src/models.json

@@ -25988,6 +25988,43 @@
 				}
 			}
 		},
+		"gemini-3.7-flash": {
+			"id": "gemini-3.7-flash",
+			"name": "Gemini 3.7 Flash",
+			"api": "google-gemini-cli",
+			"provider": "google-antigravity",
+			"baseUrl": "https://daily-cloudcode-pa.googleapis.com",
+			"reasoning": true,
+			"input": [
+				"text",
+				"image"
+			],
+			"cost": {
+				"input": 0.75,
+				"output": 3.75,
+				"cacheRead": 0.075,
+				"cacheWrite": 0
+			},
+			"contextWindow": 1048576,
+			"maxTokens": 65536,
+			"requestModelId": "gemini-3.7-flash-low",
+			"thinking": {
+				"mode": "google-level",
+				"efforts": [
+					"minimal",
+					"low",
+					"medium",
+					"high"
+				],
+				"requiresEffort": true,
+				"effortRouting": {
+					"minimal": "gemini-3.7-flash-low",
+					"low": "gemini-3.7-flash-low",
+					"medium": "gemini-3.7-flash-medium",
+					"high": "gemini-3.7-flash-high"
+				}
+			}
+		},
 		"gpt-oss-120b": {
 			"id": "gpt-oss-120b",
 			"name": "GPT OSS 120B",
@@ -29128,12 +29165,12 @@
 				"text"
 			],
 			"cost": {
-				"input": 0.079996,
-				"output": 0.252,
-				"cacheRead": 0.0252,
+				"input": 0.072,
+				"output": 0.144,
+				"cacheRead": 0.016,
 				"cacheWrite": 0
 			},
-			"contextWindow": 1048576,
+			"contextWindow": 262144,
 			"maxTokens": 262144,
 			"thinking": {
 				"mode": "effort",
@@ -29156,10 +29193,10 @@
 				"image"
 			],
 			"cost": {
-				"input": 1.5,
-				"output": 7.5,
-				"cacheRead": 0.15,
-				"cacheWrite": 0.083333
+				"input": 0.375,
+				"output": 1.875,
+				"cacheRead": 0.0375,
+				"cacheWrite": 0.020833
 			},
 			"contextWindow": 1048576,
 			"maxTokens": 65536,
@@ -29216,13 +29253,13 @@
 				"image"
 			],
 			"cost": {
-				"input": 2.8,
-				"output": 14,
-				"cacheRead": 0.29,
+				"input": 2.4,
+				"output": 12,
+				"cacheRead": 0.24,
 				"cacheWrite": 0
 			},
-			"contextWindow": 1048576,
-			"maxTokens": 1048576,
+			"contextWindow": 974842,
+			"maxTokens": 974842,
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
@@ -32370,10 +32407,10 @@
 				"image"
 			],
 			"cost": {
-				"input": 1.5,
-				"output": 7.5,
-				"cacheRead": 0.15,
-				"cacheWrite": 0.083333
+				"input": 0.75,
+				"output": 3.75,
+				"cacheRead": 0.075,
+				"cacheWrite": 0.041667
 			},
 			"contextWindow": 1048576,
 			"maxTokens": 65536,
@@ -32394,19 +32431,29 @@
 			"api": "openai-completions",
 			"provider": "kilo",
 			"baseUrl": "https://api.kilo.ai/api/gateway",
-			"reasoning": false,
+			"reasoning": true,
 			"input": [
-				"text"
+				"text",
+				"image"
 			],
 			"cost": {
-				"input": 0,
-				"output": 0,
-				"cacheRead": 0,
-				"cacheWrite": 0
+				"input": 0.75,
+				"output": 3.75,
+				"cacheRead": 0.075,
+				"cacheWrite": 0.041667
 			},
 			"contextWindow": 1048576,
 			"maxTokens": 65536,
-			"supportsComputerUse": false
+			"thinking": {
+				"mode": "effort",
+				"efforts": [
+					"minimal",
+					"low",
+					"medium",
+					"high"
+				],
+				"requiresEffort": true
+			}
 		},
 		"google/gemma-2-27b-it": {
 			"id": "google/gemma-2-27b-it",
@@ -41453,9 +41500,9 @@
 				"text"
 			],
 			"cost": {
-				"input": 0.55,
-				"output": 2.2,
-				"cacheRead": 0.11,
+				"input": 0.5,
+				"output": 2,
+				"cacheRead": 0.1,
 				"cacheWrite": 0
 			},
 			"contextWindow": 202752,
@@ -41533,9 +41580,9 @@
 				"text"
 			],
 			"cost": {
-				"input": 0.6,
-				"output": 2.2,
-				"cacheRead": 0.11,
+				"input": 0.4,
+				"output": 1.75,
+				"cacheRead": 0.08,
 				"cacheWrite": 0
 			},
 			"contextWindow": 202752,
@@ -41562,7 +41609,7 @@
 				"text"
 			],
 			"cost": {
-				"input": 0.07,
+				"input": 0.06,
 				"output": 0.4,
 				"cacheRead": 0.01,
 				"cacheWrite": 0
@@ -41591,8 +41638,8 @@
 				"text"
 			],
 			"cost": {
-				"input": 1,
-				"output": 3.2,
+				"input": 0.95,
+				"output": 2.55,
 				"cacheRead": 0.2,
 				"cacheWrite": 0
 			},
@@ -41649,7 +41696,7 @@
 				"text"
 			],
 			"cost": {
-				"input": 1.38,
+				"input": 1.4,
 				"output": 4.4,
 				"cacheRead": 0.26,
 				"cacheWrite": 0
@@ -41683,7 +41730,7 @@
 				"cacheRead": 0.26,
 				"cacheWrite": 0
 			},
-			"contextWindow": 262144,
+			"contextWindow": 1048576,
 			"maxTokens": 131072,
 			"thinking": {
 				"mode": "effort",
@@ -51050,7 +51097,7 @@
 				"input": 1.5,
 				"output": 9,
 				"cacheRead": 0.15,
-				"cacheWrite": 0
+				"cacheWrite": 0.083333
 			},
 			"contextWindow": 1048576,
 			"maxTokens": 65536,
@@ -51110,7 +51157,37 @@
 				"input": 1.5,
 				"output": 7.5,
 				"cacheRead": 0.15,
-				"cacheWrite": 0
+				"cacheWrite": 0.041667
+			},
+			"contextWindow": 1048576,
+			"maxTokens": 65536,
+			"thinking": {
+				"mode": "effort",
+				"efforts": [
+					"minimal",
+					"low",
+					"medium",
+					"high"
+				],
+				"requiresEffort": true
+			}
+		},
+		"google/gemini-3.7-flash": {
+			"id": "google/gemini-3.7-flash",
+			"name": "Gemini 3.7 Flash",
+			"api": "openai-completions",
+			"provider": "nanogpt",
+			"baseUrl": "https://nano-gpt.com/api/v1",
+			"reasoning": true,
+			"input": [
+				"text",
+				"image"
+			],
+			"cost": {
+				"input": 0.375,
+				"output": 1.875,
+				"cacheRead": 0.0375,
+				"cacheWrite": 0.020833
 			},
 			"contextWindow": 1048576,
 			"maxTokens": 65536,
@@ -71596,6 +71673,35 @@
 				]
 			}
 		},
+		"deepseek-v4-pro:preview": {
+			"id": "deepseek-v4-pro:preview",
+			"name": "deepseek-v4-pro:preview",
+			"api": "ollama-chat",
+			"provider": "ollama-cloud",
+			"baseUrl": "https://ollama.com",
+			"reasoning": true,
+			"thinking": {
+				"mode": "effort",
+				"efforts": [
+					"low",
+					"high",
+					"max"
+				]
+			},
+			"input": [
+				"text"
+			],
+			"cost": {
+				"input": 0,
+				"output": 0,
+				"cacheRead": 0,
+				"cacheWrite": 0
+			},
+			"contextWindow": 524288,
+			"maxTokens": 65536,
+			"omitMaxOutputTokens": true,
+			"supportsComputerUse": false
+		},
 		"devstral-2:123b": {
 			"id": "devstral-2:123b",
 			"name": "devstral-2:123b",
@@ -76127,6 +76233,36 @@
 				"requiresEffort": true
 			}
 		},
+		"gemini-3.7-flash": {
+			"id": "gemini-3.7-flash",
+			"name": "Gemini 3.7 Flash",
+			"api": "google-generative-ai",
+			"provider": "opencode-zen",
+			"baseUrl": "https://opencode.ai/zen/v1",
+			"reasoning": true,
+			"input": [
+				"text",
+				"image"
+			],
+			"cost": {
+				"input": 1.5,
+				"output": 7.5,
+				"cacheRead": 0.15,
+				"cacheWrite": 0
+			},
+			"contextWindow": 1048576,
+			"maxTokens": 65536,
+			"thinking": {
+				"mode": "google-level",
+				"efforts": [
+					"minimal",
+					"low",
+					"medium",
+					"high"
+				],
+				"requiresEffort": true
+			}
+		},
 		"glm-4.6": {
 			"id": "glm-4.6",
 			"name": "GLM-4.6",
@@ -77974,13 +78110,13 @@
 				"text"
 			],
 			"cost": {
-				"input": 0.079996,
-				"output": 0.252,
-				"cacheRead": 0.0252,
+				"input": 0.072,
+				"output": 0.144,
+				"cacheRead": 0.016,
 				"cacheWrite": 0
 			},
 			"contextWindow": 1048576,
-			"maxTokens": 131072,
+			"maxTokens": 262144,
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
@@ -78007,7 +78143,7 @@
 				"input": 0.375,
 				"output": 1.875,
 				"cacheRead": 0.0375,
-				"cacheWrite": 0.0416666666666667
+				"cacheWrite": 0.0208333333333333
 			},
 			"contextWindow": 1048576,
 			"maxTokens": 65536,
@@ -78066,13 +78202,13 @@
 				"image"
 			],
 			"cost": {
-				"input": 2.8,
-				"output": 14,
-				"cacheRead": 0.29,
+				"input": 2.4,
+				"output": 12,
+				"cacheRead": 0.24,
 				"cacheWrite": 0
 			},
 			"contextWindow": 1048576,
-			"maxTokens": 1048576,
+			"maxTokens": 974842,
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
@@ -81235,10 +81371,10 @@
 				"image"
 			],
 			"cost": {
-				"input": 1.5,
-				"output": 7.5,
-				"cacheRead": 0.15,
-				"cacheWrite": 0.0833333333333333
+				"input": 0.75,
+				"output": 3.75,
+				"cacheRead": 0.075,
+				"cacheWrite": 0.0416666666666667
 			},
 			"contextWindow": 1048576,
 			"maxTokens": 65536,
@@ -81267,10 +81403,10 @@
 				"image"
 			],
 			"cost": {
-				"input": 0.75,
-				"output": 3.75,
-				"cacheRead": 0.075,
-				"cacheWrite": 0.0833333333333333
+				"input": 0.375,
+				"output": 1.875,
+				"cacheRead": 0.0375,
+				"cacheWrite": 0.0416666666666667
 			},
 			"contextWindow": 1048576,
 			"maxTokens": 65536,
@@ -81302,7 +81438,7 @@
 				"input": 0.375,
 				"output": 1.875,
 				"cacheRead": 0.0375,
-				"cacheWrite": 0.0416666666666667
+				"cacheWrite": 0.0208333333333333
 			},
 			"contextWindow": 1048576,
 			"maxTokens": 65536,
@@ -81331,10 +81467,10 @@
 				"image"
 			],
 			"cost": {
-				"input": 0.375,
-				"output": 1.875,
-				"cacheRead": 0.0375,
-				"cacheWrite": 0.0416666666666667
+				"input": 0.1875,
+				"output": 0.9375,
+				"cacheRead": 0.01875,
+				"cacheWrite": 0.0208333333333333
 			},
 			"contextWindow": 1048576,
 			"maxTokens": 65536,
@@ -83285,9 +83421,9 @@
 				"image"
 			],
 			"cost": {
-				"input": 0.5795,
-				"output": 2.44,
-				"cacheRead": 0.0976,
+				"input": 0.5605,
+				"output": 2.36,
+				"cacheRead": 0.0944,
 				"cacheWrite": 0
 			},
 			"contextWindow": 262144,
@@ -83915,7 +84051,7 @@
 				"cacheRead": 0.049999999999999996,
 				"cacheWrite": 0
 			},
-			"contextWindow": 1048576,
+			"contextWindow": 1000000,
 			"maxTokens": 262144,
 			"thinking": {
 				"mode": "effort",
@@ -90620,9 +90756,9 @@
 				"text"
 			],
 			"cost": {
-				"input": 0.49,
-				"output": 1.54,
-				"cacheRead": 0.091,
+				"input": 0.392,
+				"output": 1.232,
+				"cacheRead": 0.0728,
 				"cacheWrite": 0
 			},
 			"contextWindow": 1048576,
@@ -96815,6 +96951,36 @@
 			},
 			"supportsComputerUse": false
 		},
+		"alibaba/qwen3.8-2.4t-a95b": {
+			"id": "alibaba/qwen3.8-2.4t-a95b",
+			"name": "Qwen3.8 2.4T A95B",
+			"api": "anthropic-messages",
+			"provider": "vercel-ai-gateway",
+			"baseUrl": "https://ai-gateway.vercel.sh",
+			"reasoning": true,
+			"input": [
+				"text"
+			],
+			"cost": {
+				"input": 2,
+				"output": 6,
+				"cacheRead": 0.25,
+				"cacheWrite": 0
+			},
+			"contextWindow": 262144,
+			"maxTokens": 131072,
+			"thinking": {
+				"mode": "budget",
+				"efforts": [
+					"minimal",
+					"low",
+					"medium",
+					"high",
+					"xhigh"
+				]
+			},
+			"supportsComputerUse": false
+		},
 		"alibaba/qwen3.8-max": {
 			"id": "alibaba/qwen3.8-max",
 			"name": "Qwen 3.8 Max",
@@ -98529,7 +98695,7 @@
 				"cacheRead": 0,
 				"cacheWrite": 0
 			},
-			"contextWindow": 256000,
+			"contextWindow": 262144,
 			"maxTokens": 131072,
 			"thinking": {
 				"mode": "budget",

packages/catalog/src/wire/gemini-headers.ts

@@ -15,30 +15,86 @@ export const getGeminiCliHeaders = (modelId?: string) => ({
 	"Client-Metadata": "ideType=IDE_UNSPECIFIED,platform=PLATFORM_UNSPECIFIED,pluginType=GEMINI",
 });
 
-export const ANTIGRAVITY_SYSTEM_INSTRUCTION =
-	"You are Antigravity, a powerful agentic AI coding assistant designed by the Google Deepmind team working on Advanced Agentic Coding." +
-	"You are pair programming with a USER to solve their coding task. The task may require creating a new codebase, modifying or debugging an existing codebase, or simply answering a question." +
-	"**Absolute paths only**" +
-	"**Proactiveness**";
 /**
  * Antigravity / Cloud Code Assist user agent. Lives in its own file so discovery
  * and usage code can read it without pulling the heavy google-gemini-cli provider
  * (and its @google/genai → google-auth-library dependency chain) into the startup
  * parse graph.
+ *
+ * Format captured from the real 2.8.0 `antigravity/hub` client:
+ * `antigravity/hub/2.8.0 (aidev_client; os_type=darwin; arch=arm64; cl=963137146)`.
+ * The backend gates newer models (e.g. gemini-3.7-flash) on the client version,
+ * so the version tracks the latest Antigravity release via the update manifest
+ * (see {@link ensureAntigravityVersion}) with `DEFAULT_ANTIGRAVITY_VERSION` as
+ * the offline fallback. os_type/arch are pinned to the darwin/arm64 reference
+ * client the version and manifest are captured from, independent of the host
+ * platform. Overrides: PI_AI_ANTIGRAVITY_VERSION / _CL / _OS / _ARCH.
  */
-export let getAntigravityUserAgent = () => {
-	const DEFAULT_ANTIGRAVITY_VERSION = "2.1.4";
-	const version = process.env.PI_AI_ANTIGRAVITY_VERSION || DEFAULT_ANTIGRAVITY_VERSION;
-	// Map Node.js platform/arch to Antigravity's expected format.
-	// Verified against Antigravity source: _qn() and wqn() in main.js.
-	// process.platform: win32→windows, others pass through (darwin, linux)
-	// process.arch:     x64→amd64, ia32→386, others pass through (arm64)
-	const os = process.platform === "win32" ? "windows" : process.platform;
-	const arch = process.arch === "x64" ? "amd64" : process.arch === "ia32" ? "386" : process.arch;
-	const userAgent = `antigravity/hub/${version} ${os}/${arch}`;
-	getAntigravityUserAgent = () => userAgent;
-	return userAgent;
-};
+export const DEFAULT_ANTIGRAVITY_VERSION = "2.8.0";
+
+const ANTIGRAVITY_VERSION_MANIFEST_URL =
+	"https://antigravity-hub-auto-updater-974169037036.us-central1.run.app/manifest/latest-arm64-mac.yml";
+const ANTIGRAVITY_VERSION_FETCH_TIMEOUT_MS = 5_000;
+
+let discoveredAntigravityVersion: string | null = null;
+let antigravityVersionFetch: Promise<void> | null = null;
+
+/** Current Antigravity client version: env override → manifest-discovered → pinned fallback. */
+export function getAntigravityVersion(): string {
+	return process.env.PI_AI_ANTIGRAVITY_VERSION || discoveredAntigravityVersion || DEFAULT_ANTIGRAVITY_VERSION;
+}
+
+/**
+ * Extracts the client version from an electron-builder update manifest.
+ * Returns null when no well-formed `version:` line is present.
+ */
+export function parseAntigravityManifestVersion(yamlText: string): string | null {
+	for (const line of yamlText.split(/\r?\n/)) {
+		const match = /^\s*version\s*:\s*(?:"([^"]*)"|'([^']*)'|([^\s#]+))\s*(?:#.*)?$/.exec(line);
+		if (!match) continue;
+		const version = (match[1] ?? match[2] ?? match[3] ?? "").trim();
+		return /^\d+\.\d+\.\d+$/.test(version) ? version : null;
+	}
+	return null;
+}
+
+/**
+ * Resolves the latest Antigravity release from the official update manifest.
+ * Success is cached for the process lifetime; failures are silent (the pinned
+ * fallback stays valid) and clear the in-flight cache so a later call retries.
+ * Skipped entirely when PI_AI_ANTIGRAVITY_VERSION is set.
+ */
+export function ensureAntigravityVersion(fetcher: typeof fetch = fetch, signal?: AbortSignal): Promise<void> {
+	if (process.env.PI_AI_ANTIGRAVITY_VERSION || discoveredAntigravityVersion) return Promise.resolve();
+	if (antigravityVersionFetch) return antigravityVersionFetch;
+
+	antigravityVersionFetch = (async () => {
+		try {
+			const timeoutSignal = AbortSignal.timeout(ANTIGRAVITY_VERSION_FETCH_TIMEOUT_MS);
+			const response = await fetcher(ANTIGRAVITY_VERSION_MANIFEST_URL, {
+				headers: { "Cache-Control": "no-cache", "User-Agent": "electron-builder" },
+				signal: signal ? AbortSignal.any([signal, timeoutSignal]) : timeoutSignal,
+			});
+			if (response.ok) {
+				discoveredAntigravityVersion = parseAntigravityManifestVersion(await response.text());
+			}
+		} catch {
+			// Silent: the pinned fallback remains valid when version discovery fails.
+		} finally {
+			if (!discoveredAntigravityVersion) antigravityVersionFetch = null;
+		}
+	})();
+	return antigravityVersionFetch;
+}
+
+/** Antigravity `User-Agent` header value; rebuilt when the discovered version changes. */
+export function getAntigravityUserAgent(): string {
+	const version = getAntigravityVersion();
+	const cl = process.env.PI_AI_ANTIGRAVITY_CL || "963137146";
+	const os = process.env.PI_AI_ANTIGRAVITY_OS || "darwin";
+	const arch = process.env.PI_AI_ANTIGRAVITY_ARCH || "arm64";
+	return `antigravity/hub/${version} (aidev_client; os_type=${os}; arch=${arch}; cl=${cl})`;
+}
 
 /**
  * Per-wire-id Antigravity Cloud Code Assist request constants, captured from the

packages/coding-agent/src/web/search/providers/gemini.ts

@@ -9,11 +9,7 @@
  * endpoint.
  */
 import { type AuthStorage, type FetchImpl, type OAuthAccess, withOAuthAccess } from "@oh-my-pi/pi-ai";
-import {
-	ANTIGRAVITY_SYSTEM_INSTRUCTION,
-	getAntigravityUserAgent,
-	getGeminiCliHeaders,
-} from "@oh-my-pi/pi-catalog/wire/gemini-headers";
+import { getAntigravityUserAgent, getGeminiCliHeaders } from "@oh-my-pi/pi-catalog/wire/gemini-headers";
 import { fetchWithRetry, USER_AGENT } from "@oh-my-pi/pi-utils";
 
 import type { SearchCitation, SearchResponse, SearchSource } from "../../../web/search/types";
@@ -441,10 +437,9 @@ async function callGeminiSearch(
 			};
 
 	const normalizedSystemPrompt = systemPrompt?.toWellFormed();
-	const systemInstructionParts: Array<{ text: string }> = [
-		...(auth.isAntigravity ? [{ text: ANTIGRAVITY_SYSTEM_INSTRUCTION }] : []),
-		...(normalizedSystemPrompt ? [{ text: normalizedSystemPrompt }] : []),
-	];
+	const systemInstructionParts: Array<{ text: string }> = normalizedSystemPrompt
+		? [{ text: normalizedSystemPrompt }]
+		: [];
 
 	const requestBody: Record<string, unknown> = {
 		project: auth.projectId,