feat(models): add Gemini 4 Argon (#8970)

* feat(models): add Gemini 4 Argon

* chore(models): document Gemini model ID normalization

Author: waleedlatif1

apps/docs/content/docs/workflows/blocks/agent.mdx

@@ -168,7 +168,7 @@ Live tool-call chips stream for **OpenAI, Anthropic, Azure Anthropic, Google, Ve
 | Anthropic | Summaries only — These generations omit full thinking; Sim requests summarized thinking on streaming runs. | `claude-fable-5-1`, `claude-fable-5`, `claude-sonnet-5-5`, `claude-sonnet-5`, `claude-opus-5-5`, `claude-opus-5`, `claude-opus-4-8`, `claude-opus-4-7`, `claude-opus-4-6`, `claude-sonnet-4-6`, `claude-opus-4-5`, `claude-sonnet-4-5`, `claude-haiku-5-5`, `claude-haiku-4-5` |
 | Azure OpenAI | Summaries only — Requires OpenAI organization verification; falls back to no summaries. | `azure/gpt-6.1-sol`, `azure/gpt-6-sol`, `azure/gpt-6-luna`, `azure/gpt-6-astra`, `azure/gpt-5.6-sol`, `azure/gpt-5.6-terra`, `azure/gpt-5.6-luna`, `azure/gpt-5.5`, `azure/gpt-5.4-pro`, `azure/gpt-5.4`, `azure/gpt-5.4-mini`, `azure/gpt-5.4-nano`, `azure/gpt-5.2`, `azure/gpt-5.1`, `azure/gpt-5.1-codex`, `azure/gpt-5`, `azure/gpt-5-mini`, `azure/gpt-5-nano`, `azure/o3`, `azure/o4-mini` |
 | Azure Anthropic | Summaries only — These generations omit full thinking; Sim requests summarized thinking on streaming runs. | `azure-anthropic/claude-fable-5-1`, `azure-anthropic/claude-opus-5-5`, `azure-anthropic/claude-sonnet-5-5`, `azure-anthropic/claude-haiku-5-5`, `azure-anthropic/claude-opus-5`, `azure-anthropic/claude-opus-4-8`, `azure-anthropic/claude-opus-4-7`, `azure-anthropic/claude-opus-4-6`, `azure-anthropic/claude-opus-4-5`, `azure-anthropic/claude-sonnet-5`, `azure-anthropic/claude-sonnet-4-6`, `azure-anthropic/claude-sonnet-4-5`, `azure-anthropic/claude-opus-4-1`, `azure-anthropic/claude-haiku-4-5` |
-| Google | Summaries only | `gemini-3.8-flash`, `gemini-3.7-flash`, `gemini-3.6-flash`, `gemini-3.5-flash-lite`, `gemini-3.5-flash`, `gemini-3.1-pro-preview`, `gemini-3.1-flash-lite`, `gemini-3-flash-preview`, `gemini-2.5-pro`, `gemini-2.5-flash`, `gemini-2.5-flash-lite` |
+| Google | Summaries only | `gemini-4-argon`, `gemini-3.8-flash`, `gemini-3.7-flash`, `gemini-3.6-flash`, `gemini-3.5-flash-lite`, `gemini-3.5-flash`, `gemini-3.1-pro-preview`, `gemini-3.1-flash-lite`, `gemini-3-flash-preview`, `gemini-2.5-pro`, `gemini-2.5-flash`, `gemini-2.5-flash-lite` |
 | Vertex AI | Summaries only | `vertex/gemini-3.8-flash`, `vertex/gemini-3.7-flash`, `vertex/gemini-3.6-flash`, `vertex/gemini-3.5-flash-lite`, `vertex/gemini-3.5-flash`, `vertex/gemini-3.1-pro-preview`, `vertex/gemini-3.1-flash-lite`, `vertex/gemini-3-flash-preview`, `vertex/gemini-2.5-pro`, `vertex/gemini-2.5-flash`, `vertex/gemini-2.5-flash-lite` |
 | DeepSeek | Full thinking deltas | `deepseek-v4-pro`, `deepseek-flash`, `deepseek-v4-flash`, `deepseek-reasoner` |
 | xAI | Full thinking deltas | `grok-4.6`, `grok-4.5`, `grok-4.3`, `grok-4.20-multi-agent-0309` |

apps/sim/lib/knowledge/documents/image-extraction-gemini.ts

@@ -9,7 +9,7 @@ import { resolveProviderKey } from '@/lib/embeddings/keys'
 import { PermanentDocumentProcessingError } from '@/lib/knowledge/documents/document-processing-error'
 import { readBoundedHttpErrorPayload } from '@/lib/knowledge/documents/utils'
 import { getProviderModels } from '@/providers/models'
-import { isGemini3Model } from '@/providers/utils'
+import { isGemini3OrLaterModel } from '@/providers/utils'
 
 const logger = createLogger('GeminiImageExtraction')
 
@@ -22,7 +22,7 @@ export const DEFAULT_IMAGE_EXTRACTION_MODEL = 'gemini-3.8-flash'
 
 /**
  * Transcription needs little reasoning, and Gemini bills thinking as output that
- * also spends the output budget. Every Gemini 3 model accepts `LOW`.
+ * also spends the output budget. Gemini 3 and later models accept `LOW`.
  */
 const THINKING_LEVEL = 'LOW'
 
@@ -72,19 +72,19 @@ export class GeminiImageExtractionError extends Error {
 
 /**
  * The model from `KB_IMAGE_EXTRACTION_MODEL`, or the default when it is unset or
- * not a Gemini 3 model in the catalog: the request uses Gemini 3's JSON-schema
- * output and thinking level, and only catalog models carry the pricing that
+ * not a Gemini 3 or later model in the catalog: the request uses modern Gemini
+ * JSON-schema output and thinking level, and only catalog models carry the pricing that
  * bills each image. Never throws: a bad value must not stop indexing.
  */
 export function resolveImageExtractionModel(): string {
   const configured = env.KB_IMAGE_EXTRACTION_MODEL?.trim()
   if (!configured) return DEFAULT_IMAGE_EXTRACTION_MODEL
   const catalogModel = getProviderModels('google').find(
-    (id) => id.toLowerCase() === configured.toLowerCase() && isGemini3Model(id)
+    (id) => id.toLowerCase() === configured.toLowerCase() && isGemini3OrLaterModel(id)
   )
   if (catalogModel) return catalogModel
   logger.warn(
-    `KB_IMAGE_EXTRACTION_MODEL="${configured}" is not a supported Gemini 3 model, falling back to ${DEFAULT_IMAGE_EXTRACTION_MODEL}`
+    `KB_IMAGE_EXTRACTION_MODEL="${configured}" is not a supported Gemini 3 or later model, falling back to ${DEFAULT_IMAGE_EXTRACTION_MODEL}`
   )
   return DEFAULT_IMAGE_EXTRACTION_MODEL
 }

apps/sim/providers/gemini/config.ts

@@ -0,0 +1,29 @@
+import { FunctionCallingConfigMode, type GenerateContentConfig, type Schema } from '@google/genai'
+import { cleanSchemaForGemini } from '@/providers/google/utils'
+import type { ProviderRequest } from '@/providers/types'
+import { isGemini3OrLaterModel } from '@/providers/utils'
+
+/** Applies the response schema only on turns that can accept it alongside their tool configuration. */
+export function applyGeminiResponseFormat(
+  config: GenerateContentConfig,
+  responseFormat: ProviderRequest['responseFormat'],
+  model: string
+): void {
+  if (!responseFormat) return
+
+  if (config.toolConfig?.functionCallingConfig?.mode === FunctionCallingConfigMode.ANY) {
+    config.responseMimeType = undefined
+    config.responseSchema = undefined
+    config.responseJsonSchema = undefined
+    return
+  }
+
+  if (config.tools?.length && !isGemini3OrLaterModel(model)) return
+
+  config.responseMimeType = 'application/json'
+  if (isGemini3OrLaterModel(model)) {
+    config.responseJsonSchema = responseFormat.schema
+  } else {
+    config.responseSchema = cleanSchemaForGemini(responseFormat.schema) as Schema
+  }
+}

apps/sim/providers/gemini/core.test.ts

@@ -118,6 +118,7 @@ describe('Vertex Gemini request compatibility', () => {
   })
 
   it.each([
+    'gemini-4-argon',
     'vertex/gemini-3.8-flash',
     'vertex/gemini-3.7-flash',
     'vertex/gemini-3.6-flash',

apps/sim/providers/gemini/core.ts

@@ -7,7 +7,6 @@ import {
   type GoogleGenAI,
   type Interactions,
   type Part,
-  type Schema,
   type ThinkingConfig,
   type ToolConfig,
 } from '@google/genai'
@@ -21,11 +20,11 @@ import {
   captureProviderConversationStep,
   recordProviderConversationToolError,
 } from '@/providers/conversation-history'
+import { applyGeminiResponseFormat } from '@/providers/gemini/config'
 import { createGeminiStreamingToolLoopStream } from '@/providers/gemini/streaming-tool-loop'
 import { priceGeminiTokens, splitGeminiTokens, splitGeminiUsage } from '@/providers/gemini/usage'
 import {
   checkForForcedToolUsage,
-  cleanSchemaForGemini,
   convertToGeminiFormat,
   convertUsageMetadata,
   createReadableStreamFromGeminiStream,
@@ -53,7 +52,7 @@ import type {
 } from '@/providers/types'
 import {
   isDeepResearchModel,
-  isGemini3Model,
+  isGemini3OrLaterModel,
   prepareToolExecution,
   prepareToolsWithUsageControl,
   sumToolCosts,
@@ -362,19 +361,14 @@ function buildNextConfig(
   if (allForcedToolsUsed && request.responseFormat) {
     nextConfig.tools = undefined
     nextConfig.toolConfig = undefined
-    if (isGemini3Model(model)) {
-      logger.info('Gemini 3: Stripping tools after forced tool execution, schema already set')
-    } else {
-      nextConfig.responseMimeType = 'application/json'
-      nextConfig.responseSchema = cleanSchemaForGemini(request.responseFormat.schema) as Schema
-      logger.info('Using structured output for final response after tool execution')
-    }
+    logger.info('Using structured output for final response after tool execution')
   } else if (state.currentToolConfig) {
     nextConfig.toolConfig = state.currentToolConfig
   } else {
     nextConfig.toolConfig = { functionCallingConfig: { mode: FunctionCallingConfigMode.AUTO } }
   }
 
+  applyGeminiResponseFormat(nextConfig, request.responseFormat, model)
   return nextConfig
 }
 
@@ -1006,35 +1000,20 @@ export async function executeGeminiRequest(
       geminiConfig.systemInstruction = systemInstruction
     }
 
-    // Handle response format
-    if (request.responseFormat && !tools?.length) {
-      geminiConfig.responseMimeType = 'application/json'
-      geminiConfig.responseSchema = cleanSchemaForGemini(request.responseFormat.schema) as Schema
-      logger.info('Using Gemini native structured output format')
-    } else if (request.responseFormat && tools?.length && isGemini3Model(model)) {
-      geminiConfig.responseMimeType = 'application/json'
-      geminiConfig.responseJsonSchema = request.responseFormat.schema
-      logger.info('Using Gemini 3 structured output with tools (responseJsonSchema)')
-    } else if (request.responseFormat && tools?.length) {
-      logger.warn(
-        'Gemini 2 does not support responseFormat with tools. Structured output will be applied after tool execution.'
-      )
-    }
-
-    // Gemini 3.x takes thinkingLevel directly; Gemini 2.5-series rejects it and needs thinkingBudget.
+    // Gemini 3 and later take thinkingLevel directly; Gemini 2.5-series rejects it and needs thinkingBudget.
     // includeThoughts is required for thought parts to appear; it is requested
     // only on agent-events runs so legacy runs keep the pre-agent-events payload.
     if (request.thinkingLevel && request.thinkingLevel !== 'none') {
       const thinkingConfig: ThinkingConfig = { includeThoughts: request.agentEvents === true }
-      if (isGemini3Model(model)) {
+      if (isGemini3OrLaterModel(model)) {
         thinkingConfig.thinkingLevel = mapToThinkingLevel(request.thinkingLevel)
       } else {
         thinkingConfig.thinkingBudget = mapToThinkingBudget(model, request.thinkingLevel)
       }
       geminiConfig.thinkingConfig = thinkingConfig
     } else if (
       request.thinkingLevel === 'none' &&
-      !isGemini3Model(model) &&
+      !isGemini3OrLaterModel(model) &&
       supportsDisablingGemini25Thinking(model)
     ) {
       // Omitting thinkingConfig falls back to the API's dynamic default (ON for gemini-2.5-flash),
@@ -1087,17 +1066,19 @@ export async function executeGeminiRequest(
       }
     }
 
+    applyGeminiResponseFormat(geminiConfig, request.responseFormat, model)
+
     const initialCallTime = Date.now()
     /**
      * Gemini 2 cannot combine responseSchema with tools, so structured output
      * is applied on a final schema-configured request after tools settle — the
      * silent path does this; the live loop would break as soon as a turn has
-     * no calls and skip the schema. Gemini 3 carries responseJsonSchema
-     * alongside tools, so its live loop keeps structured output.
+     * no calls and skip the schema. Gemini 3 and later carry responseJsonSchema
+     * alongside automatic tools, or apply it after forced calls settle.
      */
     const hasActiveTools = Boolean(geminiConfig.tools?.length)
     const responseFormatNeedsFinalPass =
-      Boolean(request.responseFormat) && hasActiveTools && !isGemini3Model(model)
+      Boolean(request.responseFormat) && hasActiveTools && !isGemini3OrLaterModel(model)
     const liveToolLoopSupported = !responseFormatNeedsFinalPass
     const shouldStream = request.stream && !hasActiveTools
 
@@ -1262,10 +1243,7 @@ export async function executeGeminiRequest(
         tools: undefined,
         toolConfig: undefined,
       }
-      if (request.responseFormat && !isGemini3Model(model)) {
-        finalConfig.responseMimeType = 'application/json'
-        finalConfig.responseSchema = cleanSchemaForGemini(request.responseFormat.schema) as Schema
-      }
+      applyGeminiResponseFormat(finalConfig, request.responseFormat, model)
 
       const finalStartTime = Date.now()
       const finalResponsePayload = await prepareConversationGeneration(request, 'gemini', {

apps/sim/providers/gemini/streaming-tool-loop.ts

@@ -20,7 +20,6 @@ import {
   type GenerateContentResponse,
   type GoogleGenAI,
   type Part,
-  type Schema,
   type ToolConfig,
 } from '@google/genai'
 import type { Logger } from '@sim/logger'
@@ -32,9 +31,9 @@ import {
   captureProviderConversationStep,
   recordProviderConversationToolError,
 } from '@/providers/conversation-history'
+import { applyGeminiResponseFormat } from '@/providers/gemini/config'
 import {
   checkForForcedToolUsage,
-  cleanSchemaForGemini,
   convertUsageMetadata,
   ensureStructResponse,
   geminiRetryDelayMs,
@@ -51,7 +50,7 @@ import {
 import { ensureToolCallId } from '@/providers/tool-call-id'
 import { enrichLastModelSegment } from '@/providers/trace-enrichment'
 import type { ModelPricing, ProviderRequest, TimeSegment } from '@/providers/types'
-import { isGemini3Model, prepareToolExecution, sumToolCosts } from '@/providers/utils'
+import { prepareToolExecution, sumToolCosts } from '@/providers/utils'
 import type { GeminiUsage } from './types'
 import { priceGeminiTokens, splitGeminiUsage } from './usage'
 
@@ -96,19 +95,14 @@ function buildNextConfig(
   if (allForcedToolsUsed && request.responseFormat) {
     nextConfig.tools = undefined
     nextConfig.toolConfig = undefined
-    if (isGemini3Model(model)) {
-      logger.info('Gemini 3: Stripping tools after forced tool execution, schema already set')
-    } else {
-      nextConfig.responseMimeType = 'application/json'
-      nextConfig.responseSchema = cleanSchemaForGemini(request.responseFormat.schema) as Schema
-      logger.info('Using structured output for final response after tool execution')
-    }
+    logger.info('Using structured output for final response after tool execution')
   } else if (currentToolConfig) {
     nextConfig.toolConfig = currentToolConfig
   } else {
     nextConfig.toolConfig = { functionCallingConfig: { mode: FunctionCallingConfigMode.AUTO } }
   }
 
+  applyGeminiResponseFormat(nextConfig, request.responseFormat, model)
   return nextConfig
 }
 
@@ -310,6 +304,8 @@ export function createGeminiStreamingToolLoopStream(
                   model
                 )
 
+            applyGeminiResponseFormat(turnConfig, request.responseFormat, model)
+
             const modelStart = Date.now()
             const turnPayload = await prepareConversationGeneration(request, 'gemini', {
               model,