Update configs for chatgpt-5

Author: juggler434

express-server/src/routes/create.ts

@@ -246,7 +246,7 @@ const buildPipelineJob = (
         cruxes: updatedConfig.cruxesEnabled,
       },
       llm: {
-        model: "gpt-4o-mini", // ! Change when we allow different models
+        model: "gpt-5-mini", // ! Change when we allow different models
       },
     },
     data: updatedConfig.data,

pyserver/config.py

@@ -8,7 +8,7 @@
 DRY_RUN = False
 
 # cheapest for testing
-MODEL = "gpt-4o-mini"  # prod default: "gpt-4-turbo-preview"
+MODEL = "gpt-5-mini"  # prod default: "gpt-4-turbo-preview"
 
 COST_BY_MODEL = {
     # GPT-4o mini: Input is $0.150 / 1M tokens, Output is $0.600 / 1M tokens
@@ -17,6 +17,8 @@
     # GPT-4o : Input is $2.50 / 1M tokens, Output is $10.00/1M tokens
     # 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}
 }
 
 # for web-app mode, require at least 3 words in order to extract meaningful claims

pyserver/main.py

@@ -277,7 +277,6 @@ def comments_to_tree(
             {"role": "system", "content": req.llm.system_prompt},
             {"role": "user", "content": full_prompt},
         ],
-        temperature=0.0,
         response_format={"type": "json_object"},
     )
     try:
@@ -383,7 +382,6 @@ def comment_to_claims(llm: dict, comment: str, tree: dict, api_key: str) -> dict
             },
             {"role": "user", "content": full_prompt},
         ],
-        temperature=0.0,
         response_format={"type": "json_object"},
     )
     try:
@@ -708,7 +706,6 @@ def dedup_claims(claims: list, llm: LLMConfig, api_key: str) -> dict:
             {"role": "system", "content": llm.system_prompt},
             {"role": "user", "content": full_prompt},
         ],
-        temperature=0.0,
         response_format={"type": "json_object"},
     )
     try:
@@ -1264,7 +1261,6 @@ def cruxes_for_topic(
             {"role": "system", "content": llm.system_prompt},
             {"role": "user", "content": full_prompt},
         ],
-        temperature=0.0,
         response_format={"type": "json_object"},
     )
     crux = response.choices[0].message.content