Set default model to gpt-5-mini-2025-08-07 across all components

Replaced all hardcoded gpt-4o-mini, gpt-4o, and gpt-5-nano defaults with
gpt-5-mini-2025-08-07 in: refactored_analyzers.py, research.py,
report_generator.py, app.py, evaluators.py, country_profiler.py,
pollster_profiler.py, politician_profiler.py, and main_pipeline.py.

CLI --model argument still overrides the default when specified.

https://claude.ai/code/session_012K59SFzJtkQR85Ekj32XCX

app.py

@@ -157,7 +157,7 @@ def run_analysis(url: str, force_refresh: bool = False, model: str = None):
                        "duration_seconds": round(time.time() - t0, 2),
                        "status": "success",
                        "details": {"outlet_name": outlet_name,
-                                   "model": model or "gpt-4o-mini"}})
+                                   "model": model or "gpt-5-mini-2025-08-07"}})
         progress.progress(75, text="Analysis complete")
 
         # 3. Generate report
@@ -172,7 +172,7 @@ def run_analysis(url: str, force_refresh: bool = False, model: str = None):
                        "started_at": datetime.fromtimestamp(t0, tz=timezone.utc).isoformat(),
                        "duration_seconds": round(time.time() - t0, 2),
                        "status": "success",
-                       "details": {"model": model or "gpt-4o"}})
+                       "details": {"model": model or "gpt-5-mini-2025-08-07"}})
         progress.progress(90, text="Saving results...")
 
         # 4. Save

country_profiler.py

@@ -116,7 +116,7 @@ class CountryProfiler:
     ECONOMIC_WEIGHT = 0.70
     SOCIAL_WEIGHT = 0.30
 
-    def __init__(self, model: str = "gpt-4o-mini", temperature: float = 0.0):
+    def __init__(self, model: str = "gpt-5-mini-2025-08-07", temperature: float = 0.0):
         self.llm = ChatOpenAI(model=model, temperature=temperature)
 
     def _score_to_bias_label(self, score: float) -> str:

evaluators.py

@@ -249,7 +249,7 @@ class CoTWrapper(BaseModel):
 class BaseRunner:
     """Base class for all evaluation runners."""
 
-    def __init__(self, model_name: str = "gpt-5-nano-2025-08-07"):
+    def __init__(self, model_name: str = "gpt-5-mini-2025-08-07"):
         self.model_name = model_name
         # LangChain client for structured output
         self.llm = ChatOpenAI(model=model_name, temperature=0.0)
@@ -414,7 +414,7 @@ NewsGuard, or any other bias-rating service anywhere in your response.
 class LLMArticlesRunner(BaseRunner):
     """Scenario 2: LLM + 20 scraped articles + fact-check search."""
 
-    def __init__(self, model_name: str = "gpt-5-nano-2025-08-07", max_articles: int = 20,
+    def __init__(self, model_name: str = "gpt-5-mini-2025-08-07", max_articles: int = 20,
                  article_cache=None):
         super().__init__(model_name)
         self.max_articles = max_articles
@@ -654,7 +654,7 @@ class MBCRunner(BaseRunner):
                 matching the paper's incremental expansion approach.
     """
 
-    def __init__(self, model_name: str = "gpt-5-nano-2025-08-07"):
+    def __init__(self, model_name: str = "gpt-5-mini-2025-08-07"):
         super().__init__(model_name)
         self.queries_path = os.path.join("data", "queries.json")
         # Fallback queries if file is missing (based on Appendix E of the paper)
@@ -898,7 +898,7 @@ class SystemRunner(BaseRunner):
     compliant scoring. Set benchmark_mode=False for strict methodology mode.
     """
 
-    def __init__(self, model_name: str = "gpt-5-nano-2025-08-07", article_cache=None,
+    def __init__(self, model_name: str = "gpt-5-mini-2025-08-07", article_cache=None,
                  benchmark_mode: bool = True, use_synthesis: bool = True,
                  use_calibration: bool = True):
         super().__init__(model_name)
@@ -1518,7 +1518,7 @@ class HybridRunner(SystemRunner):
     2. Always uses LLM synthesis (inherits from SystemRunner with use_synthesis=True).
     """
 
-    def __init__(self, model_name: str = "gpt-5-nano-2025-08-07", article_cache=None,
+    def __init__(self, model_name: str = "gpt-5-mini-2025-08-07", article_cache=None,
                  benchmark_mode: bool = True, use_synthesis: bool = True,
                  use_calibration: bool = True):
         # HybridRunner always uses synthesis

main_pipeline.py

@@ -95,7 +95,7 @@ if __name__ == "__main__":
     parser = argparse.ArgumentParser(description="Media Bias Analysis Pipeline")
     parser.add_argument("url", help="The URL of the news site to analyze")
     parser.add_argument("--refresh", action="store_true", help="Ignore cache and force re-analysis")
-    parser.add_argument("--model", type=str, default=None, help="LLM model to use (e.g. gpt-4o, gpt-4.1)")
+    parser.add_argument("--model", type=str, default=None, help="LLM model to use (default: gpt-5-mini-2025-08-07)")
 
     args = parser.parse_args()
 

politician_profiler.py

@@ -112,7 +112,7 @@ class PoliticianProfiler:
     Factuality: 100% from failed fact checks by IFCN-approved checkers
     """
 
-    def __init__(self, model: str = "gpt-4o-mini", temperature: float = 0.0,
+    def __init__(self, model: str = "gpt-5-mini-2025-08-07", temperature: float = 0.0,
                  search_backend=None):
         self.llm = ChatOpenAI(model=model, temperature=temperature)
         self.weights = PoliticianBiasWeights()

pollster_profiler.py

@@ -109,7 +109,7 @@ class PollsterProfiler:
     Credibility: Standard MBFC methodology (assessed separately)
     """
 
-    def __init__(self, model: str = "gpt-4o-mini", temperature: float = 0.0,
+    def __init__(self, model: str = "gpt-5-mini-2025-08-07", temperature: float = 0.0,
                  search_backend=None):
         self.llm = ChatOpenAI(model=model, temperature=temperature)
         self.weights = PollsterBiasWeights()

refactored_analyzers.py

@@ -79,7 +79,7 @@ logger = logging.getLogger(__name__)
 # =============================================================================
 
 
-def get_llm(model: str = "gpt-5-nano", temperature: float = 0.0) -> ChatOpenAI:
+def get_llm(model: str = "gpt-5-mini-2025-08-07", temperature: float = 0.0) -> ChatOpenAI:
     """
     Get a configured LLM instance.
 
@@ -154,7 +154,7 @@ Analyze the CONTENT and STYLE, not the URL or publication name."""
 
     def __init__(
         self,
-        model: str = "gpt-4o-mini",
+        model: str = "gpt-5-mini-2025-08-07",
         temperature: float = 0.0,
         max_text_chars: int = 1000,
     ):
@@ -372,7 +372,7 @@ Look for indicators like:
 
     def __init__(
         self,
-        model: str = "gpt-4o-mini",
+        model: str = "gpt-5-mini-2025-08-07",
         temperature: float = 0.0,
         tranco_path: Optional[str] = None,
         auto_download_tranco: bool = True,
@@ -863,7 +863,7 @@ For example, NYT is a NEWSPAPER even though they have a website and podcasts."""
 
     def __init__(
         self,
-        model: str = "gpt-4o-mini",
+        model: str = "gpt-5-mini-2025-08-07",
         temperature: float = 0.0,
         lookup_path: Optional[str] = None,
         search_backend=None,
@@ -1225,7 +1225,7 @@ If results are about fact checks OF specific claims MADE BY the outlet, those ar
 
     def __init__(
         self,
-        model: str = "gpt-4o-mini",
+        model: str = "gpt-5-mini-2025-08-07",
         temperature: float = 0.0,
         sites: list[str] | None = None,
         search_backend=None,
@@ -1481,7 +1481,7 @@ You must also detect **Vague Sourcing** or "Anonymous Authority".
 
     def __init__(
         self,
-        model: str = "gpt-4o-mini",
+        model: str = "gpt-5-mini-2025-08-07",
         temperature: float = 0.0,
     ):
         self.llm = get_llm(model, temperature).with_structured_output(SourcingLLMOutput)
@@ -1883,7 +1883,7 @@ If the outlet is a peer-reviewed scientific journal, academic publisher, or evid
 
     def __init__(
         self,
-        model: str = "gpt-4o-mini",
+        model: str = "gpt-5-mini-2025-08-07",
         temperature: float = 0.0,
     ):
         """
@@ -2255,7 +2255,7 @@ NONE_DETECTED: No pseudoscience content found
 
     def __init__(
         self,
-        model: str = "gpt-4o-mini",
+        model: str = "gpt-5-mini-2025-08-07",
         temperature: float = 0.0,
     ):
         """
@@ -2421,7 +2421,7 @@ If no evidence of transparency is found (no about page, no ownership disclosure,
 
     def __init__(
         self,
-        model: str = "gpt-4o-mini",
+        model: str = "gpt-5-mini-2025-08-07",
         temperature: float = 0.0,
         search_backend=None,
     ):
@@ -2605,7 +2605,7 @@ Base your assessment ONLY on the article content provided."""
 
     def __init__(
         self,
-        model: str = "gpt-4o-mini",
+        model: str = "gpt-5-mini-2025-08-07",
         temperature: float = 0.0,
     ):
         self.llm = get_llm(model, temperature).with_structured_output(OneSidednessLLMOutput)

report_generator.py

@@ -7,7 +7,7 @@ from langchain_openai import ChatOpenAI
 from schemas import ComprehensiveReportData
 
 class ReportGenerator:
-    def __init__(self, model: str = "gpt-4o", temperature: float = 0.4):
+    def __init__(self, model: str = "gpt-5-mini-2025-08-07", temperature: float = 0.4):
         self.llm = ChatOpenAI(model=model, temperature=temperature)
 
     def generate(self, data: ComprehensiveReportData) -> str:

research.py

@@ -168,7 +168,7 @@ Include up to 3-5 most relevant and credible analyses."""
 
     def __init__(
         self,
-        model: str = "gpt-4o-mini",
+        model: str = "gpt-5-mini-2025-08-07",
         temperature: float = 0.0,
         search_backend=None,
     ):
@@ -1070,7 +1070,7 @@ class MediaProfiler:
 
     def __init__(
         self,
-        model: str = "gpt-4o-mini",
+        model: str = "gpt-5-mini-2025-08-07",
         temperature: float = 0.0,
         search_backend=None,
         use_calibration: bool = True,