shortcutxl 0.3.75 → 0.3.76

package/BINARY-INVENTORY.json

@@ -1,6 +1,6 @@
 {
   "schemaVersion": 1,
-  "generatedAt": "2026-07-19T05:57:28.143Z",
+  "generatedAt": "2026-07-27T04:35:44.825Z",
   "package": "shortcutxl",
   "binaryExtensions": [
     ".dll",
@@ -11,7 +11,7 @@
   "files": [
     {
       "path": "xll/ShortcutXL.xll",
-      "sha256": "696c2e28259854eb1a0406f0f432f94c838d32ea5518a3a1f82514e6d97af016",
+      "sha256": "f0d06082d5296e5737b9e4416a62942d22afd1780c27c84304bb5b11b0ba1c8a",
       "source": "ShortcutXL native XLL build",
       "version": "package",
       "builtBy": "shortcut",
@@ -523,7 +523,7 @@
     },
     {
       "path": "xll/python/Scripts/httpx.exe",
-      "sha256": "54ace5c8b5d0aedc90fc7b9c68d593719ec35f0f67066fb3f4006442e720d4e7",
+      "sha256": "27001c85dbe85fa448ba575c9b41295349f72b8bcfe2e0805748bebe8ccc9888",
       "source": "httpx console launcher installed into embedded Python",
       "version": "see packaged httpx distribution",
       "builtBy": "third-party",
@@ -531,7 +531,7 @@
     },
     {
       "path": "xll/python/Scripts/idna.exe",
-      "sha256": "66ec7c50882b73888e3505d063057f15951a36627c4c9a58b5e9742f9777ab95",
+      "sha256": "9f01a7efccc9917f3ed7e7a41b950b8734be8c844db5a6b78014188ddb2255da",
       "source": "Python package console launcher installed into embedded Python",
       "version": "see owning Python package metadata in site-packages",
       "builtBy": "third-party",
@@ -539,7 +539,7 @@
     },
     {
       "path": "xll/python/Scripts/pip.exe",
-      "sha256": "d6daf847f1bf812774d489fde310516baee00a6c141d211aed1932429d8233e4",
+      "sha256": "8296d812e172cc359b23905d4a424185e80a9ab470770c8a247fb033b0c12554",
       "source": "pip console launcher installed into embedded Python",
       "version": "see packaged pip distribution",
       "builtBy": "third-party",
@@ -547,7 +547,7 @@
     },
     {
       "path": "xll/python/Scripts/pip3.13.exe",
-      "sha256": "d6daf847f1bf812774d489fde310516baee00a6c141d211aed1932429d8233e4",
+      "sha256": "8296d812e172cc359b23905d4a424185e80a9ab470770c8a247fb033b0c12554",
       "source": "pip console launcher installed into embedded Python",
       "version": "see packaged pip distribution",
       "builtBy": "third-party",
@@ -555,7 +555,7 @@
     },
     {
       "path": "xll/python/Scripts/pip3.exe",
-      "sha256": "d6daf847f1bf812774d489fde310516baee00a6c141d211aed1932429d8233e4",
+      "sha256": "8296d812e172cc359b23905d4a424185e80a9ab470770c8a247fb033b0c12554",
       "source": "pip console launcher installed into embedded Python",
       "version": "see packaged pip distribution",
       "builtBy": "third-party",
@@ -563,7 +563,7 @@
     },
     {
       "path": "xll/python/Scripts/pywin32_postinstall.exe",
-      "sha256": "94a2b7c198cbda23c1bff4a8780422e2f1e73fd03c1347fc39fa99625938e345",
+      "sha256": "72dfdcd5711cf7093921eb0da7c196f308b5a75471a124039401eed472b87547",
       "source": "Python package console launcher installed into embedded Python",
       "version": "see owning Python package metadata in site-packages",
       "builtBy": "third-party",
@@ -571,7 +571,7 @@
     },
     {
       "path": "xll/python/Scripts/pywin32_testall.exe",
-      "sha256": "f61ee14e2798955301beadbb45503e63e33dad398ed3f216b9bb98270a8bbe54",
+      "sha256": "8201c25e4d02041af95225057a08873831cb343767d1660cab9bc0891da1114b",
       "source": "Python package console launcher installed into embedded Python",
       "version": "see owning Python package metadata in site-packages",
       "builtBy": "third-party",

package/CHANGELOG.md

@@ -1,5 +1,9 @@
 # Changelog
 
+## [0.3.76]
+
+- **Claude Opus 5** - Claude Opus 5 is now available as ShortcutXL's high-accuracy Opus option.
+
 ## [0.3.75]
 
 - **Native desktop app** - Run `shortcut app` for a multi-session desktop experience with tools, approvals, subagents, attachments, skills, themes, and account controls.

package/dist/ai/models.d.ts

@@ -12,7 +12,7 @@ export declare function calculateCost<TApi extends Api>(model: Model<TApi>, usag
  *
  * Supported today:
  * - Shortcut's current and legacy OpenAI reasoning models
- * - Anthropic Messages API Opus 4.8 models (xhigh maps to adaptive effort "max")
+ * - Anthropic Messages API Opus models (xhigh maps to adaptive effort "max")
  */
 export declare function supportsXhigh<TApi extends Api>(model: Model<TApi>): boolean;
 /**

package/dist/ai/models.js

@@ -37,7 +37,7 @@ export function calculateCost(model, usage) {
  *
  * Supported today:
  * - Shortcut's current and legacy OpenAI reasoning models
- * - Anthropic Messages API Opus 4.8 models (xhigh maps to adaptive effort "max")
+ * - Anthropic Messages API Opus models (xhigh maps to adaptive effort "max")
  */
 export function supportsXhigh(model) {
     const modelId = model.id.startsWith('openai/') ? model.id.slice('openai/'.length) : model.id;
@@ -45,7 +45,7 @@ export function supportsXhigh(model) {
         return true;
     }
     if (model.api === 'anthropic-messages') {
-        return model.id.includes('opus-4-8') || model.id.includes('opus-4.8');
+        return model.id.toLowerCase().includes('opus');
     }
     return false;
 }

package/dist/ai/providers/anthropic.js

@@ -379,7 +379,7 @@ export const streamAnthropic = (model, context, options) => {
  * tolerated. These are the only Claude Opus/Sonnet/Fable models still served;
  * older budget-based models have been retired.
  */
-const ADAPTIVE_THINKING_MODELS = ['opus-4-8', 'sonnet-5', 'fable-5'];
+const ADAPTIVE_THINKING_MODELS = ['opus-5', 'opus-4-8', 'sonnet-5', 'fable-5'];
 /** Check if a model uses adaptive thinking. */
 function supportsAdaptiveThinking(modelId) {
     const normalized = modelId.toLowerCase();
@@ -419,8 +419,8 @@ export const streamSimpleAnthropic = (model, context, options) => {
             thinkingEnabled: false
         });
     }
-    // For Opus 4.6 and Sonnet 4.6: use adaptive thinking with effort level
-    // For older models: use budget-based thinking
+    // Current served Claude models use adaptive thinking with an effort level.
+    // Older models use budget-based thinking.
     if (supportsAdaptiveThinking(model.id)) {
         const effort = mapThinkingLevelToEffort(options.reasoning, model.id);
         return streamAnthropic(model, context, {
@@ -488,7 +488,7 @@ function buildParams(model, context, options) {
     if (context.tools) {
         params.tools = convertTools(context.tools);
     }
-    // Configure thinking mode: adaptive (Opus 4.6 and Sonnet 4.6) or budget-based (older models)
+    // Configure adaptive thinking for current models or budget-based thinking for older models.
     if (options?.thinkingEnabled && model.reasoning) {
         if (supportsAdaptiveThinking(model.id)) {
             // Adaptive thinking: Claude decides when and how much to think

package/dist/ai/providers/google-shared.d.ts

@@ -38,6 +38,8 @@ export declare function retainThoughtSignature(existing: string | undefined, inc
  * Models via Google APIs that require explicit tool call IDs in function calls/responses.
  */
 export declare function requiresToolCallId(modelId: string): boolean;
+export declare function isGemini3FlashModelId(modelId: string): boolean;
+export declare function isGemini3ProModelId(modelId: string): boolean;
 /**
  * Convert internal messages to Gemini Content[] format.
  */

package/dist/ai/providers/google-shared.js

@@ -57,6 +57,12 @@ function resolveThoughtSignature(isSameProviderAndModel, signature) {
 export function requiresToolCallId(modelId) {
     return modelId.startsWith('claude-') || modelId.startsWith('gpt-oss-');
 }
+export function isGemini3FlashModelId(modelId) {
+    return /^gemini-3(?:\.\d+)?-flash(?:-|$)/.test(modelId);
+}
+export function isGemini3ProModelId(modelId) {
+    return /^gemini-3(?:\.\d+)?-pro(?:-|$)/.test(modelId);
+}
 /**
  * Convert internal messages to Gemini Content[] format.
  */

package/dist/ai/providers/google-vertex.js

@@ -2,7 +2,7 @@ import { GoogleGenAI, ThinkingLevel } from '@google/genai';
 import { calculateCost } from '../models.js';
 import { AssistantMessageEventStream } from '../utils/event-stream.js';
 import { sanitizeSurrogates } from '../utils/sanitize-unicode.js';
-import { convertMessages, convertTools, isThinkingPart, mapStopReason, mapToolChoice, retainThoughtSignature } from './google-shared.js';
+import { convertMessages, convertTools, isGemini3FlashModelId, isGemini3ProModelId, isThinkingPart, mapStopReason, mapToolChoice, retainThoughtSignature } from './google-shared.js';
 import { buildBaseOptions, clampReasoning } from './simple-options.js';
 const API_VERSION = 'v1';
 const THINKING_LEVEL_MAP = {
@@ -235,7 +235,7 @@ export const streamSimpleGoogleVertex = (model, context, options) => {
     }
     const effort = clampReasoning(options.reasoning);
     const geminiModel = model;
-    if (isGemini3ProModel(geminiModel) || isGemini3FlashModel(geminiModel)) {
+    if (isGemini3ProModelId(geminiModel.id) || isGemini3FlashModelId(geminiModel.id)) {
         return streamGoogleVertex(model, context, {
             ...base,
             thinking: {
@@ -327,14 +327,8 @@ function buildParams(model, context, options = {}) {
     };
     return params;
 }
-function isGemini3ProModel(model) {
-    return model.id.includes('3-pro');
-}
-function isGemini3FlashModel(model) {
-    return model.id.includes('3-flash');
-}
 function getGemini3ThinkingLevel(effort, model) {
-    if (isGemini3ProModel(model)) {
+    if (isGemini3ProModelId(model.id)) {
         switch (effort) {
             case 'minimal':
             case 'low':

package/dist/ai/providers/google.js

@@ -3,7 +3,7 @@ import { getEnvApiKey } from '../env-api-keys.js';
 import { calculateCost } from '../models.js';
 import { AssistantMessageEventStream } from '../utils/event-stream.js';
 import { sanitizeSurrogates } from '../utils/sanitize-unicode.js';
-import { convertMessages, convertTools, isThinkingPart, mapStopReason, mapToolChoice, retainThoughtSignature } from './google-shared.js';
+import { convertMessages, convertTools, isGemini3FlashModelId, isGemini3ProModelId, isThinkingPart, mapStopReason, mapToolChoice, retainThoughtSignature } from './google-shared.js';
 import { buildBaseOptions, clampReasoning } from './simple-options.js';
 // Counter for generating unique tool call IDs
 let toolCallCounter = 0;
@@ -232,7 +232,7 @@ export const streamSimpleGoogle = (model, context, options) => {
     }
     const effort = clampReasoning(options.reasoning);
     const googleModel = model;
-    if (isGemini3ProModel(googleModel) || isGemini3FlashModel(googleModel)) {
+    if (isGemini3ProModelId(googleModel.id) || isGemini3FlashModelId(googleModel.id)) {
         return streamGoogle(model, context, {
             ...base,
             thinking: {
@@ -311,14 +311,8 @@ function buildParams(model, context, options = {}) {
     };
     return params;
 }
-function isGemini3ProModel(model) {
-    return model.id.includes('3-pro');
-}
-function isGemini3FlashModel(model) {
-    return model.id.includes('3-flash');
-}
 function getGemini3ThinkingLevel(effort, model) {
-    if (isGemini3ProModel(model)) {
+    if (isGemini3ProModelId(model.id)) {
         switch (effort) {
             case 'minimal':
             case 'low':

package/dist/ai/providers/provider-error-normalization.js

@@ -1,4 +1,6 @@
-import { applyModelAccessDeniedErrorDetails, isShortcutXLModelStreamErrorDetails, LLMErrorMetadataHeaders, parseStreamErrorPayload, safeModelStreamErrorMessage } from '../../contracts/model-stream.js';
+import { APIConnectionError as AnthropicAPIConnectionError } from '@anthropic-ai/sdk';
+import { APIConnectionError as OpenAIAPIConnectionError } from 'openai';
+import { applyModelAccessDeniedErrorDetails, LLMErrorMetadataHeaders, modelStreamErrorDetailsFromThrown, networkErrorDetails, parseStreamErrorPayload, providerErrorDetailsFromRawType, safeModelStreamErrorMessage, ShortcutXLModelStreamFetchStage } from '../../contracts/model-stream.js';
 import { isContextOverflowErrorMessage } from '../utils/overflow.js';
 function getErrorMessage(error) {
     return error instanceof Error ? error.message : (JSON.stringify(error) ?? String(error));
@@ -22,32 +24,57 @@ function apiV2PayloadFromHeaders(headers) {
     };
     return Object.values(payload).some((value) => value !== undefined) ? payload : undefined;
 }
-function normalizeApiV2ErrorDetails(error, rawMessage) {
-    const payload = apiV2PayloadFromHeaders(error.headers);
-    if (!payload)
-        return undefined;
+/**
+ * Classify an SDK-thrown error from what the provider itself reported.
+ * Anthropic's `APIError` carries the error type as `.type` (in-stream error
+ * events) or `.error.error.type`; OpenAI's as `.error.type` / `.code`. SDK
+ * connection failures (fetch rejected before an HTTP response) classify as
+ * network errors.
+ */
+function sdkErrorDetails(error) {
+    if (error instanceof AnthropicAPIConnectionError || error instanceof OpenAIAPIConnectionError) {
+        return networkErrorDetails(ShortcutXLModelStreamFetchStage.BeforeResponse);
+    }
+    const candidates = [
+        error.type,
+        error.error?.error?.type,
+        error.error?.type,
+        error.code,
+        error.error?.code
+    ];
+    for (const raw of candidates) {
+        if (typeof raw !== 'string')
+            continue;
+        const details = providerErrorDetailsFromRawType(raw);
+        if (details)
+            return details;
+    }
+    return undefined;
+}
+/**
+ * Build normalized details for an SDK-thrown error by merging what the
+ * provider reported on the error itself with the api-v2 gateway's metadata
+ * headers. Header values win: the gateway sets `X-LLM-Provider-Error-Type`
+ * deliberately on HTTP failures, and a real HTTP `status` beats the synthetic
+ * one the raw-type mapping supplies.
+ */
+function normalizedSdkErrorDetails(error, rawMessage) {
     const status = typeof error.status === 'number' ? error.status : undefined;
-    const details = parseStreamErrorPayload(payload, status) ?? (status ? { status } : undefined);
-    if (!details)
+    const classified = sdkErrorDetails(error);
+    const headerDetails = parseStreamErrorPayload(apiV2PayloadFromHeaders(error.headers), status);
+    if (!classified && !headerDetails)
         return undefined;
-    const accessAwareDetails = applyModelAccessDeniedErrorDetails(details, rawMessage);
+    const merged = { ...classified, ...headerDetails };
+    const accessAware = applyModelAccessDeniedErrorDetails(merged, rawMessage);
     if (isContextOverflowErrorMessage(rawMessage)) {
-        return { ...accessAwareDetails, code: 'context_overflow' };
+        return { ...accessAware, code: 'context_overflow' };
     }
-    return accessAwareDetails;
-}
-function streamDetailsFromError(error) {
-    const detailsError = error;
-    if (isShortcutXLModelStreamErrorDetails(detailsError.details))
-        return detailsError.details;
-    if (isShortcutXLModelStreamErrorDetails(detailsError.errorDetails))
-        return detailsError.errorDetails;
-    return undefined;
+    return accessAware;
 }
 export function applyNormalizedProviderError(output, error) {
     const rawMessage = getErrorMessage(error);
-    const details = streamDetailsFromError(error) ??
-        normalizeApiV2ErrorDetails(error, rawMessage);
+    const details = modelStreamErrorDetailsFromThrown(error) ??
+        normalizedSdkErrorDetails(error, rawMessage);
     if (!details) {
         output.errorMessage = rawMessage;
         return { usedStructuredDetails: false };

package/dist/app/agent-session.js

@@ -371,6 +371,7 @@ export class AgentSession {
                         charCount: deliveredContent.charCount,
                         imageCount: deliveredContent.imageCount,
                         conversationId,
+                        modelName: this.agent.state.model?.id,
                         userMessageId
                     })),
                 conversationId,
@@ -979,6 +980,7 @@ export class AgentSession {
             imageCount: currentImages?.length ?? 0,
             conversationId: this.sessionManager.getSessionId(),
             workbookDetails,
+            modelName: this.agent.state.model?.id,
             userMessageId
         });
         this._pendingImmediateRunAgentMetadata = promptMetadata;

package/dist/app/providers/model-resolver.js

@@ -37,13 +37,13 @@ export function resolveExactModelRef(availableModels, requested) {
 /** Default model IDs for each provider with a preferred initial model. */
 export const defaultModelPerProvider = {
     [SHORTCUT_PROVIDER_ID]: SHORTCUT_OPENAI_MODEL_ID,
-    anthropic: SHORTCUT_MODEL_ID.ClaudeOpus48,
+    anthropic: SHORTCUT_MODEL_ID.ClaudeOpus5,
     openai: SHORTCUT_OPENAI_MODEL_ID,
     'azure-openai-responses': SHORTCUT_OPENAI_MODEL_ID,
     google: 'gemini-2.5-pro',
     'google-vertex': 'gemini-3-pro-preview',
     openrouter: `openai/${SHORTCUT_OPENAI_MODEL_ID}`,
-    'vercel-ai-gateway': 'anthropic/claude-opus-4-8',
+    'vercel-ai-gateway': 'anthropic/claude-opus-5',
     xai: 'grok-4-fast-non-reasoning',
     groq: 'openai/gpt-oss-120b',
     cerebras: 'zai-glm-4.6',
@@ -51,7 +51,7 @@ export const defaultModelPerProvider = {
     minimax: 'MiniMax-M2.1',
     'minimax-cn': 'MiniMax-M2.1',
     huggingface: 'moonshotai/Kimi-K2.5',
-    opencode: SHORTCUT_MODEL_ID.ClaudeOpus48,
+    opencode: SHORTCUT_MODEL_ID.ClaudeOpus5,
     'kimi-coding': 'kimi-k2-thinking'
 };
 function getDefaultModelCandidatesForProvider(provider) {
@@ -59,7 +59,7 @@ function getDefaultModelCandidatesForProvider(provider) {
         return [
             SHORTCUT_OPENAI_MODEL_ID,
             SHORTCUT_MODEL_ID.ClaudeFable5,
-            SHORTCUT_MODEL_ID.ClaudeOpus48
+            SHORTCUT_MODEL_ID.ClaudeOpus5
         ];
     }
     return [defaultModelPerProvider[provider]].filter((modelId) => !!modelId);

package/dist/app/session/run-agent-observability.d.ts

@@ -8,6 +8,7 @@ export type RunAgentMetadata = {
     has_workbook_context: boolean;
     workbook_count: number;
     workbook_scope_mode: string | undefined;
+    modelName?: string;
     user_message_id?: string;
     turnId?: string;
 };
@@ -31,6 +32,7 @@ export declare function buildRunAgentMetadata(options: {
         workbooks?: string[];
         scopeMode?: string;
     };
+    modelName?: string;
     userMessageId?: string;
 }): RunAgentMetadata;
 export declare function emitRunAgent(observability: RunAgentObservability | undefined, metadata: RunAgentMetadata): void;

package/dist/app/session/run-agent-observability.js

@@ -33,6 +33,7 @@ export function buildRunAgentMetadata(options) {
         has_workbook_context: Boolean(options.workbookDetails),
         workbook_count: options.workbookDetails?.workbooks?.length ?? 0,
         workbook_scope_mode: options.workbookDetails?.scopeMode,
+        ...(options.modelName ? { modelName: options.modelName } : {}),
         ...(options.userMessageId
             ? { user_message_id: options.userMessageId, turnId: options.userMessageId }
             : {})

package/dist/app/session/session-models.js

@@ -25,12 +25,12 @@ export const THINKING_LEVELS_WITH_XHIGH = [
 const INTERACTIVE_MODELS_PER_PROVIDER = {
     shortcut: new Set([
         SHORTCUT_MODEL_ID.ClaudeFable5,
-        SHORTCUT_MODEL_ID.ClaudeOpus48,
+        SHORTCUT_MODEL_ID.ClaudeOpus5,
         SHORTCUT_OPENAI_MODEL_ID,
         SHORTCUT_MODEL_ID.Glm52Fireworks,
         SHORTCUT_MODEL_ID.Hosted
     ]),
-    anthropic: new Set([SHORTCUT_MODEL_ID.ClaudeOpus48]),
+    anthropic: new Set([SHORTCUT_MODEL_ID.ClaudeOpus5]),
     openai: new Set([SHORTCUT_OPENAI_MODEL_ID])
 };
 // ============================================================================

package/dist/app/subagents/defaults.d.ts

@@ -1,4 +1,4 @@
 export declare const DEFAULT_SUBAGENT_MODEL: "shortcut/gpt-5.6-sol";
 export declare const DEFAULT_SUBAGENT_TIMEOUT_SECONDS: number;
-export declare const TASK_MODEL_CHOICES: readonly ["shortcut/gpt-5.6-sol", "shortcut/claude-opus-4-8"];
+export declare const TASK_MODEL_CHOICES: readonly ["shortcut/gpt-5.6-sol", "shortcut/claude-opus-5"];
 //# sourceMappingURL=defaults.d.ts.map

package/dist/app/subagents/defaults.js

@@ -1,8 +1,5 @@
 import { SHORTCUT_MODEL_REF, SHORTCUT_OPENAI_MODEL_REF } from '../../model-ids.js';
 export const DEFAULT_SUBAGENT_MODEL = SHORTCUT_OPENAI_MODEL_REF;
 export const DEFAULT_SUBAGENT_TIMEOUT_SECONDS = 30 * 60;
-export const TASK_MODEL_CHOICES = [
-    DEFAULT_SUBAGENT_MODEL,
-    SHORTCUT_MODEL_REF.ClaudeOpus48
-];
+export const TASK_MODEL_CHOICES = [DEFAULT_SUBAGENT_MODEL, SHORTCUT_MODEL_REF.ClaudeOpus5];
 //# sourceMappingURL=defaults.js.map

package/dist/app/tools/cron.js

@@ -32,12 +32,13 @@ const ScheduleSchema = Type.Union([
         }))
     })
 ]);
+const MODEL_OVERRIDE_DESCRIPTION = `Model override, for example "${SHORTCUT_MODEL_REF.ClaudeOpus5}" for Opus.`;
 const JobInputSchema = Type.Object({
     name: Type.String({ description: 'Short name for the job' }),
     schedule: ScheduleSchema,
     payload: Type.Object({
         message: Type.String({ description: 'The prompt/instruction for the agent to execute' }),
-        model: Type.Optional(Type.String({ description: `Model override, e.g. "${SHORTCUT_MODEL_REF.ClaudeOpus48}"` }))
+        model: Type.Optional(Type.String({ description: MODEL_OVERRIDE_DESCRIPTION }))
     }),
     enabled: Type.Optional(Type.Boolean({ description: 'Whether the job is enabled (default: true)' })),
     description: Type.Optional(Type.String({ description: 'Longer description of what the job does' }))
@@ -49,7 +50,7 @@ const PatchSchema = Type.Object({
     schedule: Type.Optional(ScheduleSchema),
     payload: Type.Optional(Type.Object({
         message: Type.Optional(Type.String()),
-        model: Type.Optional(Type.String())
+        model: Type.Optional(Type.String({ description: MODEL_OVERRIDE_DESCRIPTION }))
     }))
 });
 const schema = Type.Object({

package/dist/app/tools/llm-analysis.js

@@ -8,6 +8,7 @@ import { Type } from '@sinclair/typebox';
 import { mkdir, readdir, readFile, stat, unlink, writeFile } from 'fs/promises';
 import { basename, extname, join } from 'path';
 import { getAgentTempDir } from '../../config.js';
+import { SHORTCUT_MODEL_ID } from '../../model-ids.js';
 import { LLM_ANALYSIS } from '../../tool-names.js';
 import { mapShortcutLlmProxyError } from '../providers/shortcut-llm-proxy-client.js';
 const TOOL_NAME = LLM_ANALYSIS;
@@ -53,10 +54,12 @@ const TEXT_EXTENSIONS = new Set([
     '.bas'
 ]);
 const MODELS = {
-    'gemini-3-flash-preview': 'gemini-3-flash-preview',
+    [SHORTCUT_MODEL_ID.Gemini3Flash]: SHORTCUT_MODEL_ID.Gemini3Flash,
+    [SHORTCUT_MODEL_ID.Gemini35FlashLite]: SHORTCUT_MODEL_ID.Gemini35FlashLite,
+    [SHORTCUT_MODEL_ID.Gemini36Flash]: SHORTCUT_MODEL_ID.Gemini36Flash,
     'claude-haiku-4-5-20251001': 'claude-haiku-4-5-20251001'
 };
-const DEFAULT_MODEL = 'gemini-3-flash-preview';
+const DEFAULT_MODEL = SHORTCUT_MODEL_ID.Gemini3Flash;
 const SYSTEM_PROMPT = `You are a document extraction assistant.
 
 Extraction rules:
@@ -76,7 +79,9 @@ Supported text types: ${[...TEXT_EXTENSIONS].join(', ')}
 Also supports: .pdf
 
 MODEL SELECTION:
-- "gemini-3-flash-preview" (default): Best for careful extraction, tables, and especially images
+- "${SHORTCUT_MODEL_ID.Gemini3Flash}" (default): Current production model for careful extraction, tables, and images
+- "${SHORTCUT_MODEL_ID.Gemini35FlashLite}": Fast, cost-efficient document parsing and structured extraction
+- "${SHORTCUT_MODEL_ID.Gemini36Flash}": Complex visual and multimodal reasoning
 - "claude-haiku-4-5-20251001": Quick scanning, structure discovery
 
 FOR EXTRACTION TASKS:
@@ -97,7 +102,7 @@ const schema = Type.Object({
         description: 'The question/instruction for the LLM about the file content.'
     }),
     model: Type.Optional(Type.String({
-        description: `Model to use: '${DEFAULT_MODEL}' (default, best for images/tables) or 'claude-haiku-4-5-20251001' (fallback/verifier)`
+        description: `Model to use: '${DEFAULT_MODEL}' (default), '${SHORTCUT_MODEL_ID.Gemini35FlashLite}' (fast, cost-efficient extraction), '${SHORTCUT_MODEL_ID.Gemini36Flash}' (complex visual reasoning), or 'claude-haiku-4-5-20251001' (fallback/verifier)`
     })),
     output_filename: Type.Optional(Type.String({
         description: "Optional output filename (e.g., 'analysis.md'). Auto-generated if omitted."

package/dist/app/tools/refresh-context.js

@@ -20,7 +20,7 @@ const schema = Type.Object({
         description: 'Progress folder for autonomous continuation. When autonomous mode is active, pass this explicitly so the runtime can inject it into the next-session prompt.'
     })),
     model: Type.Optional(Type.String({
-        description: `Optional model override for the next session. Accepts the same model patterns as CLI model selection, e.g. "${SHORTCUT_MODEL_REF.ClaudeOpus48}" or "opus".`
+        description: `Optional model override for the next session. Set model to "${SHORTCUT_MODEL_REF.ClaudeOpus5}" to use Opus.`
     })),
     thinkingLevel: Type.Optional(thinkingLevelSchema)
 });

package/dist/app/tools/take-screenshot.js

@@ -17,10 +17,11 @@
  */
 import { Type } from '@sinclair/typebox';
 import { TAKE_SCREENSHOT_SYSTEM_PROMPT, TAKE_SCREENSHOT_TOOL_DESCRIPTION } from '../../core/prompts/take-screenshot.js';
+import { SHORTCUT_MODEL_ID } from '../../model-ids.js';
 import { TAKE_SCREENSHOT } from '../../tool-names.js';
 import { mapShortcutLlmProxyError } from '../providers/shortcut-llm-proxy-client.js';
 const TOOL_NAME = TAKE_SCREENSHOT;
-const DEFAULT_VISION_MODEL = 'gemini-3-flash-preview';
+const DEFAULT_VISION_MODEL = SHORTCUT_MODEL_ID.Gemini3Flash;
 const SCREENSHOT_LLM_MAX_TOKENS = 16384;
 const SCREENSHOT_LLM_TIMEOUT_MS = 120_000;
 // ---------------------------------------------------------------------------

package/dist/app/tools/task/task.js

@@ -153,7 +153,7 @@ function buildSchema(excludeSubagents) {
             description: 'Full sandbox path to a .txt file containing the complete subagent prompt. Prefer this for long prompts, repeated/batched tasks, or many parallel agents using similar guidance; generate the files programmatically, then launch the tasks concurrently with these paths.'
         })),
         model: Type.Optional(StringEnum(TASK_MODEL_CHOICES, {
-            description: `${SHORTCUT_OPENAI_MODEL_REF} (default) is fastest; ${SHORTCUT_MODEL_REF.ClaudeOpus48} is as accurate, better at formatting, slower, and useful for an independent opinion.`
+            description: `${SHORTCUT_OPENAI_MODEL_REF} is the default and fastest option. Set model to "${SHORTCUT_MODEL_REF.ClaudeOpus5}" for the Opus option with higher accuracy and stronger formatting when extra latency is acceptable.`
         })),
         run_in_background: Type.Optional(Type.Boolean({
             description: RUN_IN_BACKGROUND_DESCRIPTION

package/dist/cli/list-models.js

@@ -1,6 +1,7 @@
 /**
  * List available models with optional fuzzy search
  */
+import { filterInteractiveSelectableModels } from '../app/session/session-models.js';
 import { fuzzyFilter } from '../tui/index.js';
 /**
  * Format a number as human-readable (e.g., 200000 -> "200K", 1000000 -> "1M")
@@ -20,7 +21,7 @@ function formatTokenCount(count) {
  * List available models, optionally filtered by search pattern
  */
 export async function listModels(modelRegistry, searchPattern) {
-    const models = modelRegistry.getAvailable();
+    const models = filterInteractiveSelectableModels(modelRegistry.getAvailable());
     if (models.length === 0) {
         console.log('No models available. Set API keys in environment variables.');
         return;