Opus 5: pricing entry + charts with the Opus 5 vanilla arm

CHANGELOG.md

@@ -9,6 +9,12 @@ See [docs/RELEASING.md](docs/RELEASING.md) for the release procedure.
 
 ## [Unreleased]
 
+### Added
+- Bundled price catalog: `claude-opus-5` ($5/$25 per 1M tokens, cache read
+  $0.50 / 1-hour cache write $10), verified against the official Anthropic
+  pricing page 2026-07-24; the other Claude entries re-verified at the same
+  date. ([#77])
+
 ### Changed
 - **Cost is now cache-aware ([ADR-014](docs/adr/014-cache-aware-cost.md)) — supersedes ADR-011's
   "as if every run were the first" formula.** `cost_usd` bills fresh input at the
@@ -632,4 +638,5 @@ Initial release under the **nasde-toolkit** name (rebrand from
 [#71]: https://github.com/NoesisVision/nasde-toolkit/pull/71
 [#73]: https://github.com/NoesisVision/nasde-toolkit/pull/73
 [#74]: https://github.com/NoesisVision/nasde-toolkit/pull/74
+[#77]: https://github.com/NoesisVision/nasde-toolkit/pull/77
 [gh-litellm-2026-04]: https://github.com/BerriAI/litellm/security/advisories/GHSA-xqmj-j6mv-4862

examples/ddd-architectural-challenges/cost_quality_plane.py

@@ -1,4 +1,4 @@
-"""Cost x quality plane for the 24-trial Fable/Opus grid (PL + EN publication PNGs).
+"""Cost x quality plane for the Fable/Opus grid + the Opus 5 vanilla arm (PL + EN PNGs).
 
 X = cache-aware run cost (ADR-014): fresh input at the full rate, cache writes at
 the 1h-cache write rate, cache reads at the cached rate, output at the output rate
@@ -26,41 +26,49 @@
     ("Opus 4.8", "vanilla"): ["GoWvUz6", "cyQyXFc", "JTgey8p", "Lozfurr"],
     ("Opus 4.8", "hint"): ["YgcbZjf", "VoktgLb", "sSUYQp4", "a6okSgZ"],
     ("Opus 4.8", "skill"): ["ZMX6Xbq", "TsjcHWY", "Bkwiqom", "F5vYATs"],
+    ("Opus 5", "vanilla"): ["i9ivEds", "m82SdTE", "Yi8S8wA", "Wipg7XN"],
+    ("Opus 5", "hint"): ["UBsUYEb", "ycb5f8S", "24pWHEZ", "bC6TqMF"],
+    ("Opus 5", "skill"): ["mvexh8R", "X8Kb8T6", "f7DZgwd", "jRUpT56"],
 }
-MODEL_ID = {"Fable 5": "claude-fable-5", "Opus 4.8": "claude-opus-4-8"}
-CODER_COLOR = {"Fable 5": "#2a78d6", "Opus 4.8": "#1baf7a"}
+MODEL_ID = {"Fable 5": "claude-fable-5", "Opus 4.8": "claude-opus-4-8", "Opus 5": "claude-opus-5"}
+# The article's judge panel is frozen at 2x Fable + 2x Opus 4.8; evals by other
+# judge models (e.g. the claude-opus-5 judge pilot) must not enter the means.
+JUDGES = {"claude-fable-5", "claude-opus-4-8"}
+CODER_COLOR = {"Fable 5": "#2a78d6", "Opus 4.8": "#1baf7a", "Opus 5": "#d97706"}
 CONFIG_MARKER = {"vanilla": "o", "hint": "^", "skill": "s"}
 INK = "#1a1a19"
 
 TEXT = {
     "pl": {
-        "title": "Koszt runu a jakość modelu domenowego — ddd-weather-discount",
+        "title": "Koszt runu a jakość rozwiązania — ddd-weather-discount",
         "xlabel": "koszt runu w USD — stawki API z rozliczeniem prompt cache",
         "ylabel": "jakość (średnia 4 ewaluacji, rubryka v2.3)",
         "coder": "model kodujący",
         "mean": "duży znacznik = średnia ramienia (n=4)",
         "outlier": "pojedynczy run za ${cost:.0f}",
         # \$ keeps matplotlib from treating $...$ pairs as mathtext
         "footnote": (
+            "jakość = pełna rubryka v2.3: dopasowanie modelu 50 pkt + granice i powściągliwość 25 pkt + jakość testów 25 pkt\n"
             "koszt = świeże wejście × stawka + zapisy cache × stawka zapisu (2×) "
             "+ odczyty cache × stawka odczytu (0.1×) + wyjście × stawka wyjścia\n"
             "stawki API z {as_of}: Fable 5 \\${fi:.0f} / \\${fo:.0f}, "
-            "Opus 4.8 \\${oi:.0f} / \\${oo:.0f} za mln tokenów"
+            "Opus 4.8 \\${oi:.0f} / \\${oo:.0f}, Opus 5 \\${o5i:.0f} / \\${o5o:.0f} za mln tokenów"
         ),
         "out": "cost_quality_plane.png",
     },
     "en": {
-        "title": "Run cost vs domain-model quality — ddd-weather-discount",
+        "title": "Run cost vs solution quality — ddd-weather-discount",
         "xlabel": "run cost in USD — API rates with prompt caching",
         "ylabel": "quality (mean of 4 evaluations, rubric v2.3)",
         "coder": "coding model",
         "mean": "large marker = arm mean (n=4)",
         "outlier": "a single ${cost:.0f} run",
         "footnote": (
+            "quality = the full v2.3 rubric: model fit 50 pts + boundaries & restraint 25 pts + test quality 25 pts\n"
             "cost = fresh input × input rate + cache writes × write rate (2×) "
             "+ cache reads × read rate (0.1×) + output × output rate\n"
             "API rates as of {as_of}: Fable 5 \\${fi:.0f} / \\${fo:.0f}, "
-            "Opus 4.8 \\${oi:.0f} / \\${oo:.0f} per MTok"
+            "Opus 4.8 \\${oi:.0f} / \\${oo:.0f}, Opus 5 \\${o5i:.0f} / \\${o5o:.0f} per MTok"
         ),
         "out": "cost_quality_plane_en.png",
     },
@@ -74,6 +82,9 @@
     ("Opus 4.8", "vanilla"): (-14, -4, "right"),
     ("Opus 4.8", "hint"): (14, -4, "left"),
     ("Opus 4.8", "skill"): (14, -4, "left"),
+    ("Opus 5", "vanilla"): (-14, 2, "right"),
+    ("Opus 5", "hint"): (14, 2, "left"),
+    ("Opus 5", "skill"): (14, -4, "left"),
 }
 
 
@@ -84,15 +95,17 @@ def load_rates() -> dict:
         "fo": raw["claude-fable-5"]["output_per_1m"],
         "oi": raw["claude-opus-4-8"]["input_per_1m"],
         "oo": raw["claude-opus-4-8"]["output_per_1m"],
-        "as_of": raw["claude-fable-5"]["as_of"],
+        "o5i": raw["claude-opus-5"]["input_per_1m"],
+        "o5o": raw["claude-opus-5"]["output_per_1m"],
+        "as_of": raw["claude-opus-5"]["as_of"],
         "by_model": {
             m: (
                 raw[m]["input_per_1m"],
                 raw[m]["output_per_1m"],
                 raw[m]["cached_input_per_1m"],
                 raw[m]["cache_write_per_1m"],
             )
-            for m in ("claude-fable-5", "claude-opus-4-8")
+            for m in ("claude-fable-5", "claude-opus-4-8", "claude-opus-5")
         },
     }
 
@@ -112,7 +125,7 @@ def collect(rates: dict) -> list[dict]:
             scores = []
             for f in sorted(trial_dir.glob("assessment_eval_*.json")):
                 d = json.loads(f.read_text())
-                if d.get("dimensions_fingerprint") == FP and d.get("evaluator_model") in MODEL_ID.values():
+                if d.get("dimensions_fingerprint") == FP and d.get("evaluator_model") in JUDGES:
                     scores.append(d["normalized_score"])
             if len(scores) != 4:
                 print(f"WARN: {trial} has {len(scores)} v2.3 evals (expected 4)")
@@ -154,22 +167,25 @@ def plane_plot(rows: list[dict], rates: dict, lang: str) -> None:
         ax.annotate(config, (mc, mq), textcoords="offset points", xytext=(dx, dy),
                     ha=ha, fontsize=9.5, color=col, fontweight="bold", zorder=6)
 
-    # per-coder trajectory through the arm means, in config order
+    # per-coder trajectory through the arm means, in config order (skip configs
+    # a coder was not run under — Opus 5 has a vanilla arm only)
     for coder in CODER_COLOR:
         pts = []
         for config in ("vanilla", "hint", "skill"):
             sub = [r for r in rows if r["coder"] == coder and r["config"] == config]
-            pts.append((sum(r["cost"] for r in sub) / len(sub), sum(r["q"] for r in sub) / len(sub)))
-        ax.plot([p[0] for p in pts], [p[1] for p in pts], color=CODER_COLOR[coder],
-                linewidth=1.1, linestyle=(0, (4, 3)), alpha=0.65, zorder=2)
+            if sub:
+                pts.append((sum(r["cost"] for r in sub) / len(sub), sum(r["q"] for r in sub) / len(sub)))
+        if len(pts) > 1:
+            ax.plot([p[0] for p in pts], [p[1] for p in pts], color=CODER_COLOR[coder],
+                    linewidth=1.1, linestyle=(0, (4, 3)), alpha=0.65, zorder=2)
 
     top = max(rows, key=lambda r: r["cost"])
     ax.annotate(t["outlier"].format(cost=top["cost"]), (top["cost"], top["q"]),
                 textcoords="offset points", xytext=(-13, -3), ha="right",
                 fontsize=8.8, color="#666", zorder=6)
 
     ax.set_xlim(0, 26)
-    ax.set_ylim(0.55, 0.92)
+    ax.set_ylim(0.55, 0.95)
     ax.set_xlabel(t["xlabel"], fontsize=10, color="#444")
     ax.set_ylabel(t["ylabel"], fontsize=10, color="#444")
     ax.tick_params(labelsize=9, colors="#444")

examples/ddd-architectural-challenges/verdict_heatmap.py

@@ -1,4 +1,4 @@
-"""Verdict heatmap for the 24-trial grid: 18 rubric checks x 6 arms (PL + EN PNGs).
+"""Verdict heatmap, grid + Opus 5 arm: 18 rubric checks x 7 arms (PL + EN PNGs).
 
 Per-check verdicts (M1-M7 / R1-R6 / T1-T5, FULL/PARTIAL/NONE) are parsed from the
 judges' free-text dimension reasoning — the eval JSON stores scores per dimension
@@ -28,10 +28,17 @@
     ("Opus 4.8", "vanilla"): ["GoWvUz6", "cyQyXFc", "JTgey8p", "Lozfurr"],
     ("Opus 4.8", "hint"): ["YgcbZjf", "VoktgLb", "sSUYQp4", "a6okSgZ"],
     ("Opus 4.8", "skill"): ["ZMX6Xbq", "TsjcHWY", "Bkwiqom", "F5vYATs"],
+    ("Opus 5", "vanilla"): ["i9ivEds", "m82SdTE", "Yi8S8wA", "Wipg7XN"],
+    ("Opus 5", "hint"): ["UBsUYEb", "ycb5f8S", "24pWHEZ", "bC6TqMF"],
+    ("Opus 5", "skill"): ["mvexh8R", "X8Kb8T6", "f7DZgwd", "jRUpT56"],
 }
 ARM_ORDER = [("Fable 5", "vanilla"), ("Fable 5", "hint"), ("Fable 5", "skill"),
-             ("Opus 4.8", "vanilla"), ("Opus 4.8", "hint"), ("Opus 4.8", "skill")]
-CODER_COLOR = {"Fable 5": "#2a78d6", "Opus 4.8": "#1baf7a"}
+             ("Opus 4.8", "vanilla"), ("Opus 4.8", "hint"), ("Opus 4.8", "skill"),
+             ("Opus 5", "vanilla"), ("Opus 5", "hint"), ("Opus 5", "skill")]
+CODER_COLOR = {"Fable 5": "#2a78d6", "Opus 4.8": "#1baf7a", "Opus 5": "#d97706"}
+# Panel frozen at 2x Fable + 2x Opus 4.8 per trial; evals by other judge models
+# (the claude-opus-5 judge pilot) share the fingerprint and must be excluded.
+JUDGES = {"claude-fable-5", "claude-opus-4-8"}
 INK = "#1a1a19"
 
 MAX = {"M1": 10, "M2": 9, "M3": 9, "M4": 7, "M5": 7, "M6": 4, "M7": 4,
@@ -86,14 +93,14 @@
 }
 TEXT = {
     "pl": {
-        "title": "Profil werdyktów rubryki v2.3 — 18 checków × 6 ramion "
+        "title": "Profil werdyktów rubryki v2.3 — 18 checków × 9 ramion "
                  "(komórka = średnia z 16 ocen: 4 triale × 4 ewaluacje)",
         "scale": "0% = wszędzie NONE · 50% = przeciętnie PARTIAL · 100% = wszędzie FULL",
         "groups": {"M": "model i kompozycja", "R": "granice i powściągliwość", "T": "jakość testów"},
         "out": "verdict_heatmap.png",
     },
     "en": {
-        "title": "Rubric v2.3 verdict profile — 18 checks × 6 arms "
+        "title": "Rubric v2.3 verdict profile — 18 checks × 9 arms "
                  "(cell = mean of 16 evaluations: 4 trials × 4 evals)",
         "scale": "0% = NONE everywhere · 50% = PARTIAL on average · 100% = FULL everywhere",
         "groups": {"M": "model & composition fit", "R": "boundaries & restraint", "T": "test quality"},
@@ -109,16 +116,21 @@
     rf"\b({ID}(?:\s*[/,]\s*{ID})*)"
     rf"(?:[:\-–—(=\s]|\b(?:scores?|stays?|is|are|remains?|at|all|falls?|drops?|to)\b)*"
     rf"({V}|full|partial|none)\b")
-# 2) verdict with explicit points: "R3 FULL(2)" — points win when they contradict the word
-P_V_POINTS = re.compile(rf"\b({ID})\s*({V})\s*\(\s*(\d+)")
+# 2) verdict with explicit points: "R3 FULL(2)", "M5 FULL(6/7)" — points win when
+#    they contradict the word; the paren must CLOSE after the number so prose
+#    counters like "FULL (0 annotations removed)" are not read as points
+P_V_POINTS = re.compile(rf"\b({ID})\s*({V})\s*\(\s*(\d+)\s*(?:/\s*\d+\s*)?\)")
 # 3) verdict-first: "PARTIAL on R5 (2)"
 P_V_FIRST = re.compile(rf"\b({V})s?\s+on\s+({ID})\b")
 # 4) verdict-first with a listed tail: "PARTIALs: M3 (...), M4 (...)" — IDs until sentence end
 P_V_LIST = re.compile(rf"\b({V})s?\s*(?:on)?:\s*([^.;]*)")
-# 5) bare points: "R5 2/4", "T1=0", "(R1 8, R2 5)"
-P_POINTS = re.compile(rf"\b({ID})\s*(?:[=:]\s*|\s+)(\d+)(?:\s*/\s*(\d+))?(?=[\s,;.)\]])")
+# 5) bare points: "R5 2/4", "T1=0", "(R1 8, R2 5)", "T1=9: ..." (trailing colon)
+P_POINTS = re.compile(rf"\b({ID})\s*(?:[=:]\s*|\s+)(\d+)(?:\s*/\s*(\d+))?(?=[\s,;.):\]])")
 # 6) "Lost 5 on M3"
 P_LOST = re.compile(rf"\b[Ll]ost\s+(\d+)\s+on\s+({ID})\b")
+# 7) unquantified deduction list: "Loses points on M3 (...) and M7 (...)" — a
+#    named deduction with no amount is a PARTIAL (a NONE is always spelled out)
+P_LOSES = re.compile(r"\b[Ll]oses?\s+points?\s+on\s+([^.;]*)")
 
 
 def classify(points: int, cid: str) -> str:
@@ -151,6 +163,9 @@ def parse_eval(d: dict) -> tuple[dict[str, str], list[str]]:
             local.setdefault(cid, classify(int(pts), cid))
         for lost, cid in P_LOST.findall(text):
             local.setdefault(cid, classify(MAX[cid] - int(lost), cid))
+        for tail in P_LOSES.findall(text):
+            for cid in re.findall(rf"\b{ID}\b", tail):
+                local.setdefault(cid, "PARTIAL")
         checks = DIM_CHECKS[name]
         for cid in checks:
             local.setdefault(cid, "FULL")  # judges enumerate deductions only
@@ -173,7 +188,7 @@ def collect() -> tuple[dict[tuple[str, str], list[dict]], list[str]]:
             (td,) = JOBS.glob(f"*/ddd-weather-discount__{t}")
             for f in sorted(td.glob("assessment_eval_*.json")):
                 d = json.loads(f.read_text())
-                if d.get("dimensions_fingerprint") != FP:
+                if d.get("dimensions_fingerprint") != FP or d.get("evaluator_model") not in JUDGES:
                     continue
                 verdicts, audit = parse_eval(d)
                 per_arm[arm].append(verdicts)
@@ -198,7 +213,7 @@ def heatmap(per_arm: dict, lang: str) -> None:
              for arm in ARM_ORDER] for c in CHECKS]
     rgba = [[CMAPS[c[0]](v) for v in row] for c, row in zip(CHECKS, grid)]
 
-    fig, ax = plt.subplots(figsize=(9.6, 8.8))
+    fig, ax = plt.subplots(figsize=(12.0, 8.8))
     fig.suptitle(t["title"], fontsize=11.5, color=INK, y=0.985)
     ax.imshow(rgba, aspect="auto")
 
@@ -222,7 +237,7 @@ def heatmap(per_arm: dict, lang: str) -> None:
 
     # white gridlines between cells; heavier breaks + side captions between M/R/T groups
     for xi in range(1, len(ARM_ORDER)):
-        lw = 3.4 if xi == 3 else 1.6
+        lw = 3.4 if xi in (3, 6) else 1.6
         ax.axvline(xi - 0.5, color="white", linewidth=lw)
     for yi in range(1, len(CHECKS)):
         ax.axhline(yi - 0.5, color="white", linewidth=1.6)
@@ -243,7 +258,7 @@ def heatmap(per_arm: dict, lang: str) -> None:
 if __name__ == "__main__":
     per_arm, flags = collect()
     n = sum(len(v) for v in per_arm.values())
-    print(f"evals parsed: {n} (expected 96)")
+    print(f"evals parsed: {n} (expected 144)")
     for fl in flags:
         print("FLAG", fl)
     for lang in ("pl", "en"):

src/nasde_toolkit/pricing.toml

@@ -32,20 +32,28 @@ cached_input_per_1m = 0.25
 as_of = "2026-06-08"
 source = "https://developers.openai.com/api/docs/pricing"
 
+[models."claude-opus-5"]
+input_per_1m = 5.0
+output_per_1m = 25.0
+cached_input_per_1m = 0.50
+cache_write_per_1m = 10.0
+as_of = "2026-07-24"
+source = "https://platform.claude.com/docs/en/about-claude/pricing"
+
 [models."claude-opus-4-8"]
 input_per_1m = 5.0
 output_per_1m = 25.0
 cached_input_per_1m = 0.50
 cache_write_per_1m = 10.0
-as_of = "2026-07-13"
+as_of = "2026-07-24"
 source = "https://platform.claude.com/docs/en/about-claude/pricing"
 
 [models."claude-fable-5"]
 input_per_1m = 10.0
 output_per_1m = 50.0
 cached_input_per_1m = 1.0
 cache_write_per_1m = 20.0
-as_of = "2026-07-13"
+as_of = "2026-07-24"
 source = "https://platform.claude.com/docs/en/about-claude/pricing"
 
 [models."claude-sonnet-4-6"]

tests/test_pricing.py

@@ -168,6 +168,7 @@ def test_layered_three_layers_compose(tmp_path: Path, empty_user_layer: Path) ->
     assert set(merged) == {
         "gpt-5.5",
         "gpt-5.4",
+        "claude-opus-5",
         "claude-fable-5",
         "claude-opus-4-8",
         "claude-sonnet-4-6",