From 9f04525c826f2bc37b65efe5dc7f405a8cfe0a84 Mon Sep 17 00:00:00 2001
From: Jeroen Ubbink <jeroen@blackbyte.nl>
Date: Sun, 5 Jul 2026 15:39:03 +0200
Subject: [PATCH] fix(upstream): reject unusable litellm prices instead of
 resolving them

A litellm cost entry that lists a model but prices both tokens at 0 (free
moderation/rerank tiers) was resolved as a real price, importing the model
enabled and served for free. Reject a both-zero (or negative) litellm hit so
the caller falls through to the next source or fails closed, mirroring the
_has_valid_pricing check the OpenRouter feed already applies.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---
 routstr/upstream/pricing_resolver.py |  6 +++++
 tests/unit/test_upstream_generic.py  | 39 ++++++++++++++++++++++++++++
 2 files changed, 45 insertions(+)

diff --git a/routstr/upstream/pricing_resolver.py b/routstr/upstream/pricing_resolver.py
index 4b06febe..80b27fc8 100644
--- a/routstr/upstream/pricing_resolver.py
+++ b/routstr/upstream/pricing_resolver.py
@@ -89,6 +89,12 @@ def _from_litellm(model_id: str) -> ResolvedPricing | None:
     completion = info.get("output_cost_per_token")
     if not isinstance(prompt, (int, float)) or not isinstance(completion, (int, float)):
         return None
+    # A both-zero entry is litellm listing a model without a real price (free
+    # moderation/rerank tiers do this) — treating 0/0 as resolved would serve
+    # the model for free. Reject it (and any negative) so the caller falls
+    # through, mirroring async_fetch_openrouter_models' _has_valid_pricing.
+    if prompt < 0 or completion < 0 or (prompt == 0 and completion == 0):
+        return None
 
     input_modalities = ["text"]
     if info.get("supports_vision"):
diff --git a/tests/unit/test_upstream_generic.py b/tests/unit/test_upstream_generic.py
index 3c312460..edc448a6 100644
--- a/tests/unit/test_upstream_generic.py
+++ b/tests/unit/test_upstream_generic.py
@@ -145,6 +145,45 @@ async def test_bare_deepseek_resolves_via_litellm() -> None:
     or_feed.assert_not_awaited()
 
 
+@pytest.mark.asyncio
+async def test_litellm_zero_price_entry_fails_closed(
+    caplog: pytest.LogCaptureFixture,
+) -> None:
+    """A litellm entry that lists a model but prices it at 0/0 (free-tier
+    moderation/rerank models do this) is not a real price — treating it as one
+    would silently serve the model for free. The resolver must reject a both-zero
+    litellm hit and fall through, so the model imports disabled, not at $0."""
+    payload = {
+        "data": [
+            {"id": "omni-moderation-latest", "object": "model", "owned_by": "openai"},
+        ]
+    }
+
+    gen_logger = logging.getLogger("routstr.upstream.generic")
+    gen_logger.addHandler(caplog.handler)
+    try:
+        with _patch_models_endpoint(payload):
+            or_feed = AsyncMock(return_value=[])
+            with patch(
+                "routstr.payment.models.async_fetch_openrouter_models", or_feed
+            ):
+                models = await GenericUpstreamProvider(
+                    base_url="http://x"
+                ).fetch_models()
+    finally:
+        gen_logger.removeHandler(caplog.handler)
+
+    model = _model_by_id(models, "omni-moderation-latest")
+    assert model.enabled is False
+    assert model.pricing.prompt == 0.0
+    assert model.pricing.completion == 0.0
+    assert any(
+        "omni-moderation-latest" in rec.getMessage()
+        for rec in caplog.records
+        if rec.levelno >= logging.WARNING
+    )
+
+
 # ---------------------------------------------------------------------------
 # OpenRouter fallback — litellm misses, OR carries a full payload
 # ---------------------------------------------------------------------------