fix(billing): a rate of zero is a price, not a missing price

The gate that decides a model cannot be priced on tokens asked
`input_rate and output_rate`, so a rate of zero read as an absent one and the
request was billed the whole reservation. The catalog serves such a model and
the router routes it — this PR says so in as many words — and then billing
charged it as if it had no price at all.

For a model that is free on both sides the reservation is the minimum-request
floor, so the overcharge is a msat. The bite is a model free on one side only:
the reservation there is the context window priced at the paid rate, charged
flat on every request no matter how few tokens it used. The catalog import
filter rejects only a both-zero price, so those models are ingested, served and
routed today.

Ask only whether each rate is usable. A zero rate now bills as what it is.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014X8RZzzbAuQCbavhFjTvJ4

routstr/payment/cost_calculation.py

@@ -270,11 +270,11 @@ async def calculate_cost(
         input_rate, output_rate, cache_read_rate, cache_creation_rate = pricing_rates
 
     # Truthiness is not the question: `NaN` and a negative rate are both truthy
-    # and sailed past this gate into the token math.
+    # and sailed past this gate into the token math, while a rate of zero is a
+    # price — free — and reading it as a missing one charged the whole
+    # reservation for a request the model serves for nothing.
     rates = (input_rate, output_rate, cache_read_rate, cache_creation_rate)
-    if not all(is_usable_rate(rate) for rate in rates) or not (
-        input_rate and output_rate
-    ):
+    if not all(is_usable_rate(rate) for rate in rates):
         logger.warning(
             "No usable token pricing — billing at flat MaxCostData. "
             "Token counts %s in the upstream response but cannot be "

tests/unit/test_pricing_rate_validation.py

@@ -92,6 +92,30 @@ async def test_unusable_token_rate_falls_back_to_max_cost(bad_rate: float) -> No
     assert cost.total_msats == 1234
 
 
+@pytest.mark.parametrize(
+    ("prompt", "completion", "expected_msats"),
+    [(0.0, 0.0, 0), (0.0, 2e-06, 1), (1e-06, 0.0, 1)],
+    ids=["free", "free-input", "free-output"],
+)
+@pytest.mark.asyncio
+async def test_a_rate_of_zero_is_billed_as_free_not_as_missing(
+    prompt: float, completion: float, expected_msats: int
+) -> None:
+    """Zero is a price, and the request must be billed on it.
+
+    The gate that decides a model has no token pricing was a truthiness test, so
+    a free rate read as an absent one and the request was charged the whole
+    reservation instead — on a model priced at zero for that side, which is a
+    price the catalog serves and the router routes.
+    """
+    model = _model(Pricing(prompt=prompt, completion=completion))
+
+    cost = await calculate_cost(_usage_response(), max_cost=1234, model_obj=model)
+
+    assert isinstance(cost, CostData)
+    assert cost.total_msats == expected_msats
+
+
 @pytest.mark.parametrize(
     "junk", [float("inf"), float("nan"), "Infinity"], ids=["inf", "nan", "inf-string"]
 )