From df20993d8b2d3c4572853811a4004df05232919d Mon Sep 17 00:00:00 2001
From: ProgrammerIn-wonderland
 <30693865+ProgrammerIn-wonderland@users.noreply.github.com>
Date: Mon, 8 Jun 2026 20:33:41 -0400
Subject: [PATCH] set azure openai as the default (#3233)

---
 .../drivers/ai-chat/ChatCompletionDriver.ts   |  2 +-
 .../providers/azure/AzureChatProvider.ts      | 20 +++++++++++++------
 2 files changed, 15 insertions(+), 7 deletions(-)

diff --git a/src/backend/drivers/ai-chat/ChatCompletionDriver.ts b/src/backend/drivers/ai-chat/ChatCompletionDriver.ts
index cb5a605c7..ae88fbe76 100644
--- a/src/backend/drivers/ai-chat/ChatCompletionDriver.ts
+++ b/src/backend/drivers/ai-chat/ChatCompletionDriver.ts
@@ -279,7 +279,7 @@ export class ChatCompletionDriver extends PuterDriver {
 
         let intendedProvider = args.provider || '';
         if (!args.model && !intendedProvider) {
-            intendedProvider = 'claude'; // default provider
+            intendedProvider = 'azure-openai'; // default provider
         }
         if (
             !args.model &&
diff --git a/src/backend/drivers/ai-chat/providers/azure/AzureChatProvider.ts b/src/backend/drivers/ai-chat/providers/azure/AzureChatProvider.ts
index d2eee5ac1..c05d0b5f9 100644
--- a/src/backend/drivers/ai-chat/providers/azure/AzureChatProvider.ts
+++ b/src/backend/drivers/ai-chat/providers/azure/AzureChatProvider.ts
@@ -51,7 +51,7 @@ export class AzureChatProvider implements IChatProvider {
      */
     #openAi: OpenAI;
 
-    #defaultModel = 'gpt-5-nano';
+    #defaultModel = 'gpt-5.4-nano';
 
     #meteringService: MeteringService;
 
@@ -220,13 +220,21 @@ export class AzureChatProvider implements IChatProvider {
 
         return OpenAiUtil.handle_completion_output({
             usage_calculator: ({ usage }) => {
+                const cachedTokens =
+                    usage.prompt_tokens_details?.cached_tokens ?? 0;
                 const trackedUsage = {
-                    prompt_tokens:
-                        (usage.prompt_tokens ?? 0) -
-                        (usage.prompt_tokens_details?.cached_tokens ?? 0),
+                    // OpenAI includes cached tokens in `prompt_tokens`, so we
+                    // subtract them out to meter the non-cached remainder. Grok
+                    // reports `prompt_tokens` already excluding cached tokens
+                    // (they're additive, not a subset) — subtracting there
+                    // underflows to a negative count, which bills a negative
+                    // (crediting) cost. Match xAI's `extractMeteredUsage` and
+                    // take Grok's `prompt_tokens` as-is.
+                    prompt_tokens: isGrok
+                        ? (usage.prompt_tokens ?? 0)
+                        : (usage.prompt_tokens ?? 0) - cachedTokens,
                     completion_tokens: usage.completion_tokens ?? 0,
-                    cached_tokens:
-                        usage.prompt_tokens_details?.cached_tokens ?? 0,
+                    cached_tokens: cachedTokens,
                 };
 
                 const costsOverrideFromModel = Object.fromEntries(