Switch gpt-4.1 to gpt-5, azure deployment names are now same as model name

Author: Veikkosuhonen

.env.template

@@ -1,8 +1,25 @@
-API_TOKEN=<API_TOKEN>
+# Toska
+API_TOKEN=
 
-AZURE_API_KEY=<AZURE_API_KEY>
-AZURE_RESOURCE=<AZURE_RESOURCE>
+# Azure
+AZURE_API_KEY=
+AZURE_RESOURCE=
 
-GPT_4O=<deployment-4o>
-GPT_4O_MINI=<deployment-4o-mini>
-GPT_41=<deployment-4.1>
+AZURE_OPENAI_EMBEDDER_MODEL=
+AZURE_OPENAI_EMBEDDER_DEPLOYMENT=
+
+# Encryption
+ENCRYPTION_KEY=
+ENCRYPTION_IV=
+
+# Laama
+LAAMA_API_TOKEN=
+LAAMA_API_URL=
+
+OLLAMA_EMBEDDER_MODEL=
+
+# S3
+S3_HOST=
+S3_ACCESS_KEY=
+S3_SECRET_ACCESS_KEY=
+S3_BUCKET=

README.md

@@ -31,13 +31,12 @@ See `compose.yaml` for local development. In short, in addition to the CC server
 
 Azure is used for OpenAI LLMs.
 
-Create an AI foundry resource (or something) and create deployment for the models you want to use (gpt4.1 for example).
+Create an AI foundry resource (or something) and create deployment for the models you want to use (gpt-5 for example). Always set the deployment name to the acual model name. So for model gpt-5, the deployment name should be gpt-5.
 
 Then populate .env with the following:
 ```
 AZURE_RESOURCE=<name-of-the-resource-you-created>
 AZURE_API_KEY=<asd>
-GPT_41=<name-of-the-deployment-you-created>
 ```
 
 ### S3

src/client/components/ChatV2/ChatV2.tsx

@@ -268,7 +268,7 @@ export const ChatV2 = () => {
         setActiveModel(defaultCourseModel ?? courseModels[0])
       }
     } else {
-      allowedModels = validModels.map((m) => m.name) // [gpt-4.1, gpt-4o, gpt-4o-mini, mock] 23.7.2025
+      allowedModels = validModels.map((m) => m.name) // [gpt-5, gpt-4o, gpt-4o-mini, mock] 23.7.2025
     }
 
     // Mock model is only visible to admins in production

src/config.ts

@@ -20,30 +20,29 @@ export const DEFAULT_RESET_CRON = process.env.DEFAULT_RESET_CRON || '0 0 1 */3 *
 
 export const EMBED_MODEL = process.env.EMBED_MODEL ?? 'text-embedding-small'
 
+/**
+ * name: the acual model name, which is shown to users, configures the model to be used and is also the azure deployment name.
+ */
 export const validModels = [
   {
     name: 'gpt-4o',
-    deployment: process.env.GPT_4O || '',
     context: 128_000,
   },
   {
     name: 'gpt-4o-mini',
-    deployment: process.env.GPT_4O_MINI || '',
     context: 128_000,
   },
   {
-    name: 'gpt-4.1',
-    deployment: process.env.GPT_41 || '',
+    name: 'gpt-5',
     context: 128_000,
   },
   {
     name: 'mock',
-    deployment: 'mock',
     context: 128_000,
   },
 ]
 
-export const DEFAULT_MODEL_ON_ENABLE = 'gpt-4.1'
+export const DEFAULT_MODEL_ON_ENABLE = 'gpt-5'
 
 export const DEFAULT_ASSISTANT_INSTRUCTIONS = '' // 11th August 2025 we decided it should be empty
 export const DEFAULT_MODEL_TEMPERATURE = 0.5

src/server/routes/testUtils.ts

@@ -120,7 +120,7 @@ router.post('/completions-api', async (req, res) => {
         content: 'Hello!, please explain the concept of artificial intelligence.',
       },
     ],
-    model: 'gpt-4.1',
+    model: 'gpt-5',
     options: {
       temperature: 0.9,
     },

src/server/services/langchain/chat.ts

@@ -17,19 +17,19 @@ import { MockModel } from './MockModel'
 type ChatModel = Runnable<BaseLanguageModelInput, AIMessageChunk, BaseChatModelCallOptions>
 
 const getChatModel = (model: string, tools: StructuredTool[], temperature: number): ChatModel => {
-  const deploymentName = validModels.find((m) => m.name === model)?.deployment
-  if (!deploymentName) {
+  const modelConfig = validModels.find((m) => m.name === model)
+  if (!modelConfig) {
     throw new Error(`Invalid model: ${model}`)
   }
 
   const chatModel =
-    deploymentName === 'mock'
+    modelConfig.name === 'mock'
       ? new MockModel({ tools, temperature })
       : new AzureChatOpenAI({
           model,
           azureOpenAIApiKey: AZURE_API_KEY,
           azureOpenAIApiVersion: '2023-05-15',
-          azureOpenAIApiDeploymentName: deploymentName,
+          azureOpenAIApiDeploymentName: model, // In Azure, always use the acual model name as the deployment name
           azureOpenAIApiInstanceName: AZURE_RESOURCE,
           temperature,
         }).bindTools(tools)

src/server/util/oldAzureClient.ts

@@ -44,14 +44,14 @@ const getMockCompletionEvents: () => Promise<EventStream<ChatCompletions>> = asy
 }
 
 export const getCompletionEvents = async ({ model, messages, options }: AzureOptions) => {
-  const deploymentId = validModels.find((m) => m.name === model)?.deployment
+  const modelConfig = validModels.find((m) => m.name === model)
 
-  if (!deploymentId) throw new Error(`Invalid model: ${model}, not one of ${validModels.map((m) => m.name).join(', ')}`)
+  if (!modelConfig) throw new Error(`Invalid model: ${model}, not one of ${validModels.map((m) => m.name).join(', ')}`)
 
-  if (deploymentId === 'mock') return getMockCompletionEvents()
+  if (modelConfig.name === 'mock') return getMockCompletionEvents()
 
   try {
-    const events = await oldClient.streamChatCompletions(deploymentId, messages, options)
+    const events = await oldClient.streamChatCompletions(modelConfig.name, messages, options)
 
     return events
   } catch (error: any) {

src/server/util/util.ts

@@ -22,12 +22,12 @@ export const getAllowedModels = (model: string): string[] => {
 
   // Logic: allowed models are selected by the pricing of the model
   // gpt-4o is the most expensive, so it is allowed for all
-  // gpt-4.1 is cheaper, so all models cheaper than it are allowed
+  // gpt-5 is cheaper, so all models cheaper than it are allowed
   // pricings: https://azure.microsoft.com/en-us/pricing/details/cognitive-services/openai-service/?cdn=disable
 
   if (model === 'gpt-4o') return allModels
 
-  if (model === 'gpt-4.1') return ['gpt-4.1', 'gpt-4o-mini']
+  if (model === 'gpt-5') return ['gpt-5', 'gpt-4o-mini']
 
   if (model === 'mock') return ['mock']