social-app/backend/tests/unit/services/test_llm_pricing_service.py

from __future__ import annotations

import pytest

from services.llm_pricing.service import LlmPricingService


def test_calculate_cost_uses_first_qwen_tier() -> None:
    service = LlmPricingService()

    cost = service.calculate_cost(
        model="qwen3.5-flash",
        prompt_tokens=100_000,
        completion_tokens=1_000,
        cached_prompt_tokens=10_000,
    )

    assert cost == pytest.approx(0.0202)


def test_calculate_cost_uses_second_qwen_tier() -> None:
    service = LlmPricingService()

    cost = service.calculate_cost(
        model="qwen3.5-flash",
        prompt_tokens=200_000,
        completion_tokens=5_000,
        cached_prompt_tokens=20_000,
    )

    assert cost == pytest.approx(0.1856)


def test_build_usage_metadata_calculates_cost_from_usage_summary() -> None:
    service = LlmPricingService()

    metadata = service.build_usage_metadata(
        model="qwen3.5-flash",
        usage_summary={
            "input_tokens": 2000,
            "output_tokens": 100,
            "latency_ms": 321,
            "cached_prompt_tokens": 500,
        },
    )

    assert metadata["model"] == "qwen3.5-flash"
    assert metadata["inputTokens"] == 2000
    assert metadata["outputTokens"] == 100
    assert metadata["totalTokens"] == 2100
    assert metadata["cachedPromptTokens"] == 500
    assert metadata["promptCacheHitTokens"] == 500
    assert metadata["promptCacheMissTokens"] == 1500
    assert metadata["reasoningTokens"] == 0
    assert metadata["cost"] == pytest.approx(0.00051)
    assert metadata["costSource"] == "catalog_fallback"
    assert metadata["usageComplete"] is True
    assert metadata["latencyMs"] == 321


def test_build_usage_metadata_prefers_provider_direct_cost() -> None:
    service = LlmPricingService()

    metadata = service.build_usage_metadata(
        model="deepseek-chat",
        usage_summary={
            "input_tokens": 1000,
            "output_tokens": 100,
            "latency_ms": 100,
            "cached_prompt_tokens": 0,
            "direct_cost": 0.1234,
            "direct_cost_observed": 1,
            "direct_cost_complete": 1,
        },
    )

    assert metadata["cost"] == pytest.approx(0.1234)
    assert metadata["costSource"] == "provider"
    assert metadata["usageComplete"] is True


def test_build_usage_metadata_falls_back_when_provider_cost_incomplete() -> None:
    service = LlmPricingService()

    metadata = service.build_usage_metadata(
        model="deepseek-chat",
        usage_summary={
            "input_tokens": 1000,
            "output_tokens": 100,
            "latency_ms": 100,
            "cached_prompt_tokens": 0,
            "direct_cost": 0.1234,
            "direct_cost_observed": 1,
            "direct_cost_complete": 0,
        },
    )

    assert metadata["cost"] == pytest.approx(0.0023)
    assert metadata["costSource"] == "catalog_fallback_incomplete_provider_cost"


def test_build_usage_metadata_marks_incomplete_usage_fallback() -> None:
    service = LlmPricingService()

    metadata = service.build_usage_metadata(
        model="deepseek-chat",
        usage_summary={
            "input_tokens": 0,
            "output_tokens": 0,
            "latency_ms": 0,
            "cached_prompt_tokens": 0,
            "model_call_records": 1,
            "usage_records": 0,
        },
    )

    assert metadata["costSource"] == "incomplete_usage_fallback"
    assert metadata["usageComplete"] is False