Skip to content

Enhanced ChatGPT Clone: Features Agents, MCP, Skills, DeepSeek, Anthropic, AWS, OpenAI, Responses API, Azure, Groq, o1, GPT-5, Mistral, OpenRouter, Vertex AI, Gemini, Artifacts, AI model switching, message search, Code Interpreter, langchain, DALL-E-3, OpenAPI Actions, Functions, Secure Multi-User Auth, Presets, open-source for self-hosting. Active

🧮 chore: Update Gemma Context Token Defaults (#13410)

Danny Avilacb6bd71ab9c5
diff --git a/api/utils/tokens.spec.js b/api/utils/tokens.spec.js
index a97694a872..e2c4ac9ba4 100644
--- a/api/utils/tokens.spec.js
+++ b/api/utils/tokens.spec.js
@@ -356,6 +356,37 @@ describe('getModelMaxTokens', () => {
);
});
+ test('should return correct context tokens for Gemma models', () => {
+ expect(maxTokensMap[EModelEndpoint.google].gemma).toBe(32768);
+ expect(getModelMaxTokens('gemma', EModelEndpoint.google)).toBe(
+ maxTokensMap[EModelEndpoint.google].gemma,
+ );
+ expect(getModelMaxTokens('gemma-2-9b-it', EModelEndpoint.google)).toBe(
+ maxTokensMap[EModelEndpoint.google]['gemma-2'],
+ );
+ expect(getModelMaxTokens('gemma-3-27b-it', EModelEndpoint.google)).toBe(
+ maxTokensMap[EModelEndpoint.google]['gemma-3-27b'],
+ );
+ expect(getModelMaxTokens('gemma4:latest', EModelEndpoint.google)).toBe(
+ maxTokensMap[EModelEndpoint.google].gemma4,
+ );
+ expect(getModelMaxTokens('gemma4:e4b', EModelEndpoint.google)).toBe(
+ maxTokensMap[EModelEndpoint.google].gemma4,
+ );
+ expect(getModelMaxTokens('Gemma4:31B', EModelEndpoint.custom)).toBe(
+ maxTokensMap[EModelEndpoint.custom]['gemma4:31b'],
+ );
+ expect(getModelMaxTokens('ollama/gemma4:31b', EModelEndpoint.custom)).toBe(
+ maxTokensMap[EModelEndpoint.custom]['gemma4:31b'],
+ );
+ expect(getModelMaxTokens('google/gemma-4-31B-it', EModelEndpoint.google)).toBe(
+ maxTokensMap[EModelEndpoint.google]['gemma-4-31b'],
+ );
+ expect(getModelMaxTokens('google/gemma-4-26B-A4B-it', EModelEndpoint.google)).toBe(
+ maxTokensMap[EModelEndpoint.google]['gemma-4-26b-a4b'],
+ );
+ });
+
test('should return correct tokens for partial match - Cohere models', () => {
expect(getModelMaxTokens('command', EModelEndpoint.custom)).toBe(
maxTokensMap[EModelEndpoint.custom]['command'],
diff --git a/packages/api/src/utils/tokens.ts b/packages/api/src/utils/tokens.ts
index 7e5cb53fff..fadec1ffe2 100644
--- a/packages/api/src/utils/tokens.ts
+++ b/packages/api/src/utils/tokens.ts
@@ -95,10 +95,19 @@ const cohereModels = {
const googleModels = {
/* Max I/O is combined so we subtract the amount from max response tokens for actual total */
- gemma: 8196,
+ gemma: 32768,
'gemma-2': 32768,
'gemma-3': 32768,
'gemma-3-27b': 131072,
+ 'gemma4:31b': 256000,
+ 'gemma4-31b': 256000,
+ 'gemma-4-31b': 256000,
+ 'gemma4:26b': 256000,
+ 'gemma4-26b': 256000,
+ 'gemma-4-26b-a4b': 256000,
+ 'gemma-4-26b': 256000,
+ gemma4: 128000,
+ 'gemma-4': 128000,
gemini: 30720, // -2048 from max
'gemini-pro-vision': 12288,
'gemini-1.5': 1000000,