feat(economy): verified OpenAI rates — the ledger's first real est_saved_usd (dogfood cycle 5)

gpt-4.1 (2.0, 8.0) and gpt-4.1-mini (0.4, 1.6) per MTok, live-verified
2026-06-12 (OpenAI pricing page + two independent trackers), recorded
in MODEL-ECONOMY.md §1 as the canonical refresh source. Prefix pricing
is now longest-first so versioned ids price by most-specific match.
On the repo's real telemetry: 8/8 calls priced, est 0.0126 vs naive
0.3122 USD — est_saved_usd 0.2996 (96%), measured not claimed.
272 tests green.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Author: claude

docs/product/MODEL-ECONOMY.md

@@ -14,6 +14,8 @@
 | Claude Sonnet 4.6 | $3 / $15 | Anthropic's explicit "default for most coding" |
 | Claude Opus 4.8 | $5 / $25 | per-TURN cost is "several times" Sonnet (deeper-thinking turns consume more tokens — multiplier unpublished, measure locally) |
 | Claude Fable 5 | $10 / $50 | tokenizer emits ~30–35% MORE tokens for the same content → effective per-content cost exceeds the 2× sticker vs Opus. Reserve strictly. |
+| gpt-4.1 | $2 / $8 | live-verified 2026-06-12 |
+| gpt-4.1-mini | $0.40 / $1.60 | live-verified 2026-06-12 |
 | Codex GPT-5.5 | 125/12.5/750 credits per MTok | ~5–45 credits per typical task |
 | Codex GPT-5.4 | 62.5/6.25/375 credits | Sonnet-analog on the Codex side |
 | Codex GPT-5.4-Mini | 18.75/1.875/113 credits | ~6.7× cheaper input than 5.5 — Codex's own haiku-class |

prd_taskmaster/economy.py

@@ -23,6 +23,8 @@
     "claude-sonnet-4-6": (3.0, 15.0),
     "claude-opus-4-8": (5.0, 25.0),
     "claude-fable-5": (10.0, 50.0),
+    "gpt-4.1": (2.0, 8.0),
+    "gpt-4.1-mini": (0.4, 1.6),
 }
 
 NAIVE_BASELINE_MODEL = "claude-fable-5"
@@ -125,7 +127,7 @@ def _price_key_for_model(model):
         return None
     if model in PRICES_PER_MTOK:
         return model
-    for price_key in PRICES_PER_MTOK:
+    for price_key in sorted(PRICES_PER_MTOK, key=len, reverse=True):
         if model.startswith(price_key + "-"):
             return price_key
     return None

tests/core/test_economy_report.py

@@ -1,10 +1,13 @@
 """T7: economy-report — telemetry summary per (op_class, model)."""
 
 import json
+from pathlib import Path
 
 import pytest
 
-from prd_taskmaster.economy import append_telemetry, summarize_telemetry
+from prd_taskmaster.economy import PRICES_PER_MTOK, append_telemetry, summarize_telemetry
+
+REPO_ROOT = Path(__file__).resolve().parents[2]
 
 
 def _rows():
@@ -44,6 +47,13 @@ def test_summarize_skips_malformed_lines(tmp_path):
     assert rep["total_calls"] == 1 and rep["skipped_lines"] == 1
 
 
+def test_model_economy_docs_include_verified_openai_price_rows():
+    content = (REPO_ROOT / "docs" / "product" / "MODEL-ECONOMY.md").read_text()
+
+    assert "| gpt-4.1 | $2 / $8 | live-verified 2026-06-12 |" in content
+    assert "| gpt-4.1-mini | $0.40 / $1.60 | live-verified 2026-06-12 |" in content
+
+
 def test_summarize_costs_priced_unpriced_unknown_and_corrupt(tmp_path):
     rows = [
         {
@@ -105,6 +115,50 @@ def test_summarize_costs_priced_unpriced_unknown_and_corrupt(tmp_path):
     assert costs["priced_coverage"] == pytest.approx(0.5)
 
 
+def test_verified_openai_price_math_from_prd(tmp_path):
+    f = tmp_path / "telemetry.jsonl"
+    f.write_text(json.dumps({
+        "op_class": "structured_gen",
+        "model": "gpt-4.1-mini",
+        "exit": 0,
+        "wall_ms": 100,
+        "tokens_in": 384,
+        "tokens_out": 1240,
+        "escalated": False,
+    }) + "\n")
+
+    rep = summarize_telemetry(f)
+
+    assert PRICES_PER_MTOK["gpt-4.1"] == (2.0, 8.0)
+    assert PRICES_PER_MTOK["gpt-4.1-mini"] == (0.4, 1.6)
+    costs = rep["costs"]
+    assert costs["est_cost_usd"] == pytest.approx(0.0021376)
+    assert costs["naive_cost_usd"] == pytest.approx(0.06584)
+    assert costs["est_saved_usd"] == pytest.approx(0.0637024)
+    assert costs["priced_calls"] == 1
+    assert costs["unpriced_calls"] == 0
+    assert costs["priced_coverage"] == pytest.approx(1.0)
+
+
+def test_versioned_openai_model_id_prices_by_most_specific_prefix(tmp_path):
+    f = tmp_path / "telemetry.jsonl"
+    f.write_text(json.dumps({
+        "op_class": "structured_gen",
+        "model": "gpt-4.1-mini-2025-04-14",
+        "exit": 0,
+        "wall_ms": 100,
+        "tokens_in": 384,
+        "tokens_out": 1240,
+        "escalated": False,
+    }) + "\n")
+
+    costs = summarize_telemetry(f)["costs"]
+
+    assert costs["est_cost_usd"] == pytest.approx(0.0021376)
+    assert costs["priced_calls"] == 1
+    assert costs["unpriced_calls"] == 0
+
+
 def test_append_telemetry_returns_row_reference(tmp_path):
     f = tmp_path / "telemetry.jsonl"
     first = {