Skip to content

Compress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server.

fix(savings): don't fabricate output savings for a free (zero-priced) model (#2298)

Abhay Singhec12e1818685
diff --git a/headroom/proxy/savings_tracker.py b/headroom/proxy/savings_tracker.py
index fbf36822f..5f5bfa599 100644
--- a/headroom/proxy/savings_tracker.py
+++ b/headroom/proxy/savings_tracker.py
@@ -238,7 +238,12 @@ def _estimate_output_savings_usd(model: str, tokens_saved: int) -> float:
resolved = _resolve_litellm_model(model)
info = litellm.model_cost.get(resolved, {})
output_cost_per_token = info.get("output_cost_per_token")
- if not output_cost_per_token:
+ # Distinguish "price unknown" (missing key -> fall back to the estimate)
+ # from a model that is legitimately free (output_cost_per_token == 0.0).
+ # `if not ...` treated a real 0.0 as unavailable and billed the fallback
+ # rate -> phantom output savings for a model that costs nothing. Mirrors
+ # the fix already applied to `_estimate_compression_savings_usd`.
+ if output_cost_per_token is None:
raise RuntimeError("output cost unavailable")
return float(tokens_saved) * float(output_cost_per_token)
except Exception:
diff --git a/tests/test_savings_tracker_zero_price.py b/tests/test_savings_tracker_zero_price.py
index 3fc3e91c3..8ac1975e2 100644
--- a/tests/test_savings_tracker_zero_price.py
+++ b/tests/test_savings_tracker_zero_price.py
@@ -15,8 +15,10 @@ import types
from headroom.proxy import savings_tracker as st
from headroom.proxy.savings_tracker import (
DEFAULT_FALLBACK_INPUT_COST_PER_TOKEN,
+ DEFAULT_FALLBACK_OUTPUT_COST_PER_TOKEN,
_estimate_compression_savings_usd,
_estimate_input_cost_usd,
+ _estimate_output_savings_usd,
)
@@ -61,3 +63,30 @@ def test_input_cost_zero_for_free_model(monkeypatch):
lambda: _fake_litellm({"free-model": {"input_cost_per_token": 0.0}}),
)
assert _estimate_input_cost_usd("free-model", 500_000) == 0.0
+
+
+def test_output_savings_zero_for_free_model(monkeypatch):
+ # output_cost_per_token == 0.0 (free model) must yield $0, not the fallback.
+ monkeypatch.setattr(
+ st,
+ "_get_litellm_module",
+ lambda: _fake_litellm({"free-model": {"output_cost_per_token": 0.0}}),
+ )
+ assert _estimate_output_savings_usd("free-model", 1_000_000) == 0.0
+
+
+def test_output_savings_falls_back_for_unknown_model(monkeypatch):
+ # Model absent from litellm → output_cost_per_token is None → fall back.
+ monkeypatch.setattr(st, "_get_litellm_module", lambda: _fake_litellm({}))
+ got = _estimate_output_savings_usd("unknown-model", 1_000_000)
+ assert got == 1_000_000 * DEFAULT_FALLBACK_OUTPUT_COST_PER_TOKEN
+
+
+def test_output_savings_uses_real_price_for_paid_model(monkeypatch):
+ price = 15.0 / 1_000_000
+ monkeypatch.setattr(
+ st,
+ "_get_litellm_module",
+ lambda: _fake_litellm({"paid-model": {"output_cost_per_token": price}}),
+ )
+ assert _estimate_output_savings_usd("paid-model", 1_000_000) == 1_000_000 * price