fix: use gpt-4o for synthesis step (gpt-4o-mini has 8K token limit too small) (#516)
The synthesis prompt needs ~15-20K tokens input (press context + historical context + continuity). gpt-4o-mini has an 8K token limit on GitHub Models which caused Step 1 to fail, falling back to single-prompt. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Juan Manuel Servera committed
Jun 16, 2026 at 19:48 UTC
463553993990b4b8aac1008f1385e077a87e8b2f
1 file changed
+1
-1
scripts/analyze_fallback.py
+1
-1
@@ -47,7 +47,7 @@ COMPACTED_PRESS_CONTEXT_CHARS = 14_000
47
COMPACTED_HISTORICAL_CONTEXT_CHARS = 12_000
48
SYNTHESIS_MAX_TOKENS = 2_000
49
SYNTHESIS_PROMPT_TOKEN_BUDGET = 20_000
50
-DEFAULT_SYNTHESIS_MODEL = "openai/gpt-4o-mini"
50
+DEFAULT_SYNTHESIS_MODEL = "openai/gpt-4o"
51
RETRYABLE_STATUS_CODES = {429, 500, 502, 503, 504}
52
NON_RETRYABLE_STATUS_CLASSES = {400, 401, 403, 404}
53
MAX_RETRIES = 3