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 Deepseek Pricing, Google Safety Settings (#5409)

Danny Avila87383fec2711
diff --git a/api/app/clients/GoogleClient.js b/api/app/clients/GoogleClient.js
index 5601e0e3cd..8fce279bf1 100644
--- a/api/app/clients/GoogleClient.js
+++ b/api/app/clients/GoogleClient.js
@@ -886,7 +886,8 @@ class GoogleClient extends BaseClient {
}
getSafetySettings() {
- const isGemini2 = this.modelOptions.model.includes('gemini-2.0');
+ const model = this.modelOptions.model;
+ const isGemini2 = model.includes('gemini-2.0') && !model.includes('thinking');
const mapThreshold = (value) => {
if (isGemini2 && value === 'BLOCK_NONE') {
return 'OFF';
diff --git a/api/models/tx.js b/api/models/tx.js
index 918899cacb..9554894ddf 100644
--- a/api/models/tx.js
+++ b/api/models/tx.js
@@ -96,6 +96,8 @@ const tokenValues = Object.assign(
'claude-': { prompt: 0.8, completion: 2.4 },
'command-r-plus': { prompt: 3, completion: 15 },
'command-r': { prompt: 0.5, completion: 1.5 },
+ 'deepseek-reasoner': { prompt: 0.14, completion: 0.55 },
+ deepseek: { prompt: 0.07, completion: 0.28 },
/* cohere doesn't have rates for the older command models,
so this was from https://artificialanalysis.ai/models/command-light/providers */
command: { prompt: 0.38, completion: 0.38 },
diff --git a/api/models/tx.spec.js b/api/models/tx.spec.js
index 238ca7b895..0492a0eedb 100644
--- a/api/models/tx.spec.js
+++ b/api/models/tx.spec.js
@@ -263,6 +263,37 @@ describe('AWS Bedrock Model Tests', () => {
});
});
+describe('Deepseek Model Tests', () => {
+ const deepseekModels = ['deepseek-chat', 'deepseek-coder', 'deepseek-reasoner'];
+
+ it('should return the correct prompt multipliers for all models', () => {
+ const results = deepseekModels.map((model) => {
+ const valueKey = getValueKey(model);
+ const multiplier = getMultiplier({ valueKey, tokenType: 'prompt' });
+ return tokenValues[valueKey].prompt && multiplier === tokenValues[valueKey].prompt;
+ });
+ expect(results.every(Boolean)).toBe(true);
+ });
+
+ it('should return the correct completion multipliers for all models', () => {
+ const results = deepseekModels.map((model) => {
+ const valueKey = getValueKey(model);
+ const multiplier = getMultiplier({ valueKey, tokenType: 'completion' });
+ return tokenValues[valueKey].completion && multiplier === tokenValues[valueKey].completion;
+ });
+ expect(results.every(Boolean)).toBe(true);
+ });
+
+ it('should return the correct prompt multipliers for reasoning model', () => {
+ const model = 'deepseek-reasoner';
+ const valueKey = getValueKey(model);
+ expect(valueKey).toBe(model);
+ const multiplier = getMultiplier({ valueKey, tokenType: 'prompt' });
+ const result = tokenValues[valueKey].prompt && multiplier === tokenValues[valueKey].prompt;
+ expect(result).toBe(true);
+ });
+});
+
describe('getCacheMultiplier', () => {
it('should return the correct cache multiplier for a given valueKey and cacheType', () => {
expect(getCacheMultiplier({ valueKey: 'claude-3-5-sonnet', cacheType: 'write' })).toBe(
diff --git a/api/server/services/Endpoints/google/llm.js b/api/server/services/Endpoints/google/llm.js
index 92eca9a6ab..f19d0539c7 100644
--- a/api/server/services/Endpoints/google/llm.js
+++ b/api/server/services/Endpoints/google/llm.js
@@ -87,7 +87,8 @@ function getLLMConfig(credentials, options = {}) {
maxRetries: 2,
};
- const isGemini2 = llmConfig.model.includes('gemini-2.0');
+ /** Used only for Safety Settings */
+ const isGemini2 = llmConfig.model.includes('gemini-2.0') && !llmConfig.model.includes('thinking');
const isGenerativeModel = llmConfig.model.includes('gemini');
const isChatModel = !isGenerativeModel && llmConfig.model.includes('chat');
const isTextModel = !isGenerativeModel && !isChatModel && /code|text/.test(llmConfig.model);
diff --git a/api/utils/tokens.js b/api/utils/tokens.js
index 68ceca797f..72151fcb19 100644
--- a/api/utils/tokens.js
+++ b/api/utils/tokens.js
@@ -82,7 +82,8 @@ const anthropicModels = {
};
const deepseekModels = {
- deepseek: 127500,
+ 'deepseek-reasoner': 63000, // -1000 from max (API)
+ deepseek: 63000, // -1000 from max (API)
};
const metaModels = {
diff --git a/api/utils/tokens.spec.js b/api/utils/tokens.spec.js
index b1f37bb1f4..f478c4769b 100644
--- a/api/utils/tokens.spec.js
+++ b/api/utils/tokens.spec.js
@@ -385,8 +385,15 @@ describe('Meta Models Tests', () => {
});
test('should match Deepseek model variations', () => {
- expect(getModelMaxTokens('deepseek-chat')).toBe(127500);
- expect(getModelMaxTokens('deepseek-coder')).toBe(127500);
+ expect(getModelMaxTokens('deepseek-chat')).toBe(
+ maxTokensMap[EModelEndpoint.openAI]['deepseek'],
+ );
+ expect(getModelMaxTokens('deepseek-coder')).toBe(
+ maxTokensMap[EModelEndpoint.openAI]['deepseek'],
+ );
+ expect(getModelMaxTokens('deepseek-reasoner')).toBe(
+ maxTokensMap[EModelEndpoint.openAI]['deepseek-reasoner'],
+ );
});
});