Update

tests/unit/services/openai_gateway/usage.py

@@ -1,13 +1,28 @@
 # retoor <retoor@molodetz.nl>
 
+from dataclasses import replace
+from datetime import datetime, timezone
+
 from devplacepy.services.openai_gateway.usage import (
     USAGE_FIELDS,
+    Pricing,
     accumulate_usage,
+    compute_cost,
     new_usage_totals,
+    normalize_usage,
     parse_usage_headers,
     usage_metric_cards,
 )
 
+BASE_PRICING = Pricing(
+    chat_cache_hit_per_m=0.0028,
+    chat_cache_miss_per_m=0.14,
+    chat_output_per_m=0.28,
+    vision_input_per_m=0.0,
+    vision_output_per_m=0.0,
+    embed_input_per_m=0.01,
+)
+
 GATEWAY_HEADERS = {
     "X-Gateway-Cost-USD": "0.00010000",
     "X-Gateway-Model": "molodetz",
@@ -89,3 +104,118 @@ def test_usage_metric_cards_labels_and_formatting():
     assert by_label["Total cost"] == "$0.0492"
     assert by_label["Avg cost/call"] == "$0.012300"
     assert by_label["Avg latency"] == "4200ms"
+
+
+def test_compute_cost_matches_deepseek_flat_rates_when_no_tier_configured():
+    norm = normalize_usage(
+        {
+            "prompt_tokens": 2000,
+            "completion_tokens": 100,
+            "prompt_cache_hit_tokens": 500,
+            "prompt_cache_miss_tokens": 1500,
+        }
+    )
+    total, input_cost, output_cost, native = compute_cost({}, norm, BASE_PRICING, "chat")
+    assert native is False
+    assert abs(input_cost - (500 / 1e6 * 0.0028 + 1500 / 1e6 * 0.14)) < 1e-12
+    assert abs(output_cost - (100 / 1e6 * 0.28)) < 1e-12
+    assert abs(total - (input_cost + output_cost)) < 1e-12
+
+
+def test_compute_cost_switches_to_tier2_above_threshold():
+    tiered = replace(
+        BASE_PRICING,
+        chat_cache_hit_per_m_tier2=0.005,
+        chat_cache_miss_per_m_tier2=0.25,
+        chat_output_per_m_tier2=0.5,
+        context_tier_threshold_tokens=1000,
+    )
+    below = normalize_usage({"prompt_tokens": 900, "completion_tokens": 50})
+    above = normalize_usage({"prompt_tokens": 1001, "completion_tokens": 50})
+    _, below_input, below_output, _ = compute_cost({}, below, tiered, "chat")
+    _, above_input, above_output, _ = compute_cost({}, above, tiered, "chat")
+    assert abs(below_output - (50 / 1e6 * 0.28)) < 1e-12
+    assert abs(above_output - (50 / 1e6 * 0.5)) < 1e-12
+    assert below_input != above_input
+
+
+def test_compute_cost_tier2_leaves_unset_component_at_tier1():
+    tiered = replace(
+        BASE_PRICING,
+        chat_output_per_m_tier2=0.5,
+        context_tier_threshold_tokens=100,
+    )
+    norm = normalize_usage({"prompt_tokens": 200, "completion_tokens": 10})
+    _, input_cost, output_cost, _ = compute_cost({}, norm, tiered, "chat")
+    assert abs(output_cost - (10 / 1e6 * 0.5)) < 1e-12
+    assert abs(input_cost - (200 / 1e6 * 0.14)) < 1e-12
+
+
+def test_compute_cost_applies_off_peak_discount():
+    discounted = replace(
+        BASE_PRICING,
+        off_peak_start_minute=60,
+        off_peak_end_minute=120,
+        off_peak_discount_pct=50.0,
+    )
+    norm = normalize_usage({"prompt_tokens": 1000, "completion_tokens": 100})
+    in_window = datetime(2026, 1, 1, 1, 30, tzinfo=timezone.utc)
+    outside_window = datetime(2026, 1, 1, 10, 0, tzinfo=timezone.utc)
+    total_in, _, _, _ = compute_cost({}, norm, discounted, "chat", now=in_window)
+    total_out, _, _, _ = compute_cost({}, norm, discounted, "chat", now=outside_window)
+    total_flat, _, _, _ = compute_cost({}, norm, BASE_PRICING, "chat")
+    assert abs(total_out - total_flat) < 1e-12
+    assert abs(total_in - total_flat / 2) < 1e-12
+
+
+def test_compute_cost_off_peak_wraps_past_midnight():
+    wrapped = replace(
+        BASE_PRICING,
+        off_peak_start_minute=23 * 60,
+        off_peak_end_minute=60,
+        off_peak_discount_pct=100.0,
+    )
+    norm = normalize_usage({"prompt_tokens": 1000, "completion_tokens": 100})
+    just_after_midnight = datetime(2026, 1, 1, 0, 30, tzinfo=timezone.utc)
+    midday = datetime(2026, 1, 1, 12, 0, tzinfo=timezone.utc)
+    total_wrapped, _, _, _ = compute_cost(
+        {}, norm, wrapped, "chat", now=just_after_midnight
+    )
+    total_midday, _, _, _ = compute_cost({}, norm, wrapped, "chat", now=midday)
+    assert total_wrapped == 0.0
+    assert total_midday > 0.0
+
+
+def test_compute_cost_image_branch_flat_per_call():
+    image_pricing = replace(BASE_PRICING, image_per_call=0.04)
+    norm = normalize_usage({})
+    total, input_cost, output_cost, native = compute_cost(
+        {}, norm, image_pricing, "image"
+    )
+    assert native is False
+    assert output_cost == 0.0
+    assert abs(total - 0.04) < 1e-9
+    assert abs(input_cost - 0.04) < 1e-9
+
+
+def test_compute_cost_image_native_cost_preferred():
+    from devplacepy.services.openai_gateway.usage import extract_image_usage
+
+    image_pricing = replace(BASE_PRICING, image_per_call=0.04)
+    norm = normalize_usage({})
+    usage = extract_image_usage({"usage": {"cost": 0.055}})
+    total, _, _, native = compute_cost(usage, norm, image_pricing, "image")
+    assert native is True
+    assert total == 0.055
+
+
+def test_compute_cost_native_upstream_cost_unaffected_by_tiering():
+    tiered = replace(
+        BASE_PRICING,
+        chat_output_per_m_tier2=100.0,
+        context_tier_threshold_tokens=1,
+    )
+    norm = normalize_usage({"prompt_tokens": 1000, "completion_tokens": 100})
+    total, _, _, native = compute_cost({"cost": 0.05}, norm, tiered, "chat")
+    assert native is True
+    assert total == 0.05