Update app.py

app.py

@@ -41,10 +41,12 @@ from utils import norm_issn
 logger = logging.getLogger(__name__)
 
 # ── Constants ──────────────────────────────────────────────────────────────
-MODEL       = "gpt-4o-mini"    # primary model — journal analysis, chat, repo finder
-MODEL_LIGHT = "gpt-4.1-nano"  # lightweight calls — subject normalise, cover letter
-# Tier 4 limits: gpt-4o-mini = 10M TPM, gpt-4.1-nano = 10M TPM — both very high
-# The 429s were caused by $10 monthly spend cap, not rate limits
+# Model routing — all separate quota buckets at Tier 4
+MODEL       = "gpt-4.1-mini"  # journal analysis, chat, repo finder — 10M TPM, Tier 4
+MODEL_LIGHT = "gpt-4.1-nano"  # subject browse, cover letter, license — 10M TPM, Tier 4
+# gpt-4.1-mini: $0.04/1M input (4x cheaper than gpt-4o-mini, same quality for JSON tasks)
+# gpt-4.1-nano: $0.01/1M input (ultra-cheap for simple normalisation calls)
+# Both have completely separate quota from gpt-4o-mini
 
 # ── KU APC agreement publishers (Option B: publisher + top 15% check) ────────
 # KU has agreements with 6 publishers. Badge shown if publisher matches AND