From 5262896d627fb9752242952bc217999ea28d49bf Mon Sep 17 00:00:00 2001
From: Cesar Garcia <128240629+Chesars@users.noreply.github.com>
Date: Sat, 13 Dec 2025 23:54:44 -0300
Subject: [PATCH] fix(perplexity): use API-provided cost instead of manual
 calculation (#17887)

Fixes #15337

Perplexity API returns pre-calculated costs in `usage.cost.total_cost`
that include the `request_cost` (fixed per-request fee). LiteLLM was
ignoring this and calculating costs manually, resulting in ~27x
underreporting (e.g., $0.0002 vs actual $0.006).

Changes:
- Use `usage.cost.total_cost` from Perplexity response when available
- Fall back to manual calculation if cost object not present
- Add tests for both behaviors
---
 litellm/llms/perplexity/cost_calculator.py    | 11 ++++
 .../test_perplexity_cost_calculator.py        | 60 ++++++++++++++++++-
 2 files changed, 70 insertions(+), 1 deletion(-)

diff --git a/litellm/llms/perplexity/cost_calculator.py b/litellm/llms/perplexity/cost_calculator.py
index c8fd2a682a8..463d897901b 100644
--- a/litellm/llms/perplexity/cost_calculator.py
+++ b/litellm/llms/perplexity/cost_calculator.py
@@ -20,6 +20,17 @@ def cost_per_token(model: str, usage: Usage) -> Tuple[float, float]:
     Returns:
         Tuple[float, float] - prompt_cost_in_usd, completion_cost_in_usd
     """
+    ## USE PRE-CALCULATED COST FROM PERPLEXITY IF AVAILABLE
+    ## Perplexity returns accurate cost in usage.cost.total_cost including request fees
+    cost_info = getattr(usage, "cost", None)
+    if cost_info is not None and isinstance(cost_info, dict):
+        total_cost = cost_info.get("total_cost")
+        if total_cost is not None:
+            # Return total cost as completion_cost (prompt_cost=0) since Perplexity
+            # doesn't break down by input/output in their cost object
+            return (0.0, float(total_cost))
+
+    ## FALLBACK: Calculate cost manually if Perplexity doesn't provide it
     ## GET MODEL INFO
     model_info = get_model_info(model=model, custom_llm_provider="perplexity")
 
diff --git a/tests/test_litellm/llms/perplexity/test_perplexity_cost_calculator.py b/tests/test_litellm/llms/perplexity/test_perplexity_cost_calculator.py
index f9a52100070..7fda731038a 100644
--- a/tests/test_litellm/llms/perplexity/test_perplexity_cost_calculator.py
+++ b/tests/test_litellm/llms/perplexity/test_perplexity_cost_calculator.py
@@ -370,4 +370,62 @@ class TestPerplexityCostCalculator:
         
         # Ensure costs are non-negative
         assert prompt_cost >= 0
-        assert completion_cost >= 0 
\ No newline at end of file
+        assert completion_cost >= 0
+
+    def test_uses_perplexity_provided_cost_when_available(self):
+        """
+        Test that when Perplexity provides pre-calculated cost in usage.cost.total_cost,
+        it is used directly instead of manual calculation.
+
+        This is the fix for issue #15337 - Perplexity returns accurate costs including
+        request_cost (fixed per-request fee) that LiteLLM cannot calculate.
+        """
+        # Create usage with Perplexity's cost object (as returned by the API)
+        usage = Usage(
+            prompt_tokens=100,
+            completion_tokens=50,
+            total_tokens=150
+        )
+
+        # Add the cost object that Perplexity returns
+        usage.cost = {
+            "input_tokens_cost": 0.0,
+            "output_tokens_cost": 0.002,
+            "request_cost": 0.006,
+            "total_cost": 0.008
+        }
+
+        prompt_cost, completion_cost = perplexity_cost_per_token(
+            model="sonar-pro",
+            usage=usage
+        )
+
+        # When Perplexity provides total_cost, we use it directly
+        # prompt_cost should be 0, completion_cost should be total_cost
+        assert prompt_cost == 0.0
+        assert completion_cost == 0.008
+        assert prompt_cost + completion_cost == 0.008
+
+    def test_falls_back_to_manual_calculation_when_no_cost_provided(self):
+        """
+        Test that manual cost calculation is used when Perplexity doesn't
+        provide the cost object (fallback behavior).
+        """
+        usage = Usage(
+            prompt_tokens=100,
+            completion_tokens=50,
+            total_tokens=150
+        )
+        # No cost object - should use manual calculation
+
+        prompt_cost, completion_cost = perplexity_cost_per_token(
+            model="sonar-deep-research",
+            usage=usage
+        )
+
+        # Should calculate manually: 100 * 2e-6 + 50 * 8e-6
+        expected_prompt = 100 * 2e-6
+        expected_completion = 50 * 8e-6
+
+        assert math.isclose(prompt_cost, expected_prompt, rel_tol=1e-6)
+        assert math.isclose(completion_cost, expected_completion, rel_tol=1e-6)
\ No newline at end of file