Adds in support for gpt-5-mini

Author: juggler434

express-server/src/pipeline/retryConfig.ts

@@ -34,12 +34,12 @@ export const TEST_RETRY_OPTIONS = {
  */
 export const HTTP_TIMEOUTS = {
   // 2 minutes per request - LLM processing can take time
-  REQUEST_TIMEOUT: 120000,
-  HEADERS_TIMEOUT: 120000,
-  BODY_TIMEOUT: 120000,
-  KEEP_ALIVE_TIMEOUT: 30000,
+  REQUEST_TIMEOUT: 12000000,
+  HEADERS_TIMEOUT: 12000000,
+  BODY_TIMEOUT: 12000000,
+  KEEP_ALIVE_TIMEOUT: 3000000,
   // Total operation timeout to prevent indefinite hangs (8 minutes with retries)
-  OPERATION_TIMEOUT: 480000,
+  OPERATION_TIMEOUT: 48000000,
 } as const;
 
 /**

pyserver/config.py

@@ -18,7 +18,7 @@
     # or: input is $0.0025/1K tokens, output is $0.01/1K tokens
     "gpt-4o": {"in_per_1K": 0.0025, "out_per_1K": 0.01},
     # GPT-5-mini: Input is $0.250 / 1M tokens, Output is $2.00/1M tokens
-    "gpt-5-mini": {"in_per_1K": 0.00025, "out_per_1k": 0.002}
+    "gpt-5-mini": {"in_per_1K": 0.00025, "out_per_1K": 0.002}
 }
 
 # for web-app mode, require at least 3 words in order to extract meaningful claims

pyserver/test_pipeline.py

@@ -53,7 +53,7 @@
 # NOTE: gpt-4o-mini is cheaper/better for basic tests, but it fails on some very basic deduplication
 API_KEY = os.getenv('OPENAI_API_KEY')
 base_llm = {
-  "model_name" : "gpt-4o-mini",
+  "model_name" : "gpt-5-mini",
   "system_prompt": config.SYSTEM_PROMPT,
   "api_key" : API_KEY
 }

pyserver/utils.py

@@ -21,6 +21,7 @@ def token_cost(model_name:str, tok_in:int, tok_out:int):
   if model_name not in config.COST_BY_MODEL:
     print("model undefined!")
     return -1
+
   return 0.001 * (tok_in  *  config.COST_BY_MODEL[model_name]["in_per_1K"] + tok_out * config.COST_BY_MODEL[model_name]["out_per_1K"])
 
 def cute_print(json_obj):