diff --git a/CLAUDE.md b/CLAUDE.md
index c36d4cb..834bab8 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -73,6 +73,20 @@ Pre-commit hooks may modify files (e.g. ruff format); re-`git add` if a hook rep
   fast option 10x quicker but similarly error-prone; full data, a parked "fast draft tier" idea, and the privacy
   implications of any multi-provider routing are in issue #31. Don't propose down-tiering the default model
   without rerunning that eval.
+- **Haiku 5.5, Opus 5.5, Fable 5.1: evaluated 2026-10-10, Sonnet 5.5 stays.** Production prompt, 21 cases,
+  Opus 5 adaptive reviewer, assembly cases only, 3 runs per config unless noted. Sonnet 5.5: 43/48 correct (3
+  were reviewer failures, not wrong answers), 2 errors, mean 5.2s, p95 9.4s, $0.011/call.
+  - *Haiku 5.5 with `thinking: disabled`* (accepted, unlike Sonnet 5.5): 36/48, 14 errors; mean 3.1s, p95 5.2s,
+    $0.0005/call (20x cheaper). The errors are failures to trace: misreading control flow (a straight-line
+    `.L4` called a loop), wrong register arithmetic, a hallucinated `imul eax, edi, edi`, "-O0 doesn't
+    constant-fold". It also missed a `je` targeting its own fall-through in about 5 of 12 samples; Sonnet,
+    Opus 5.5 and Fable 5.1 caught it every time.
+  - *Haiku 5.5 adaptive (the API default)*: 41/48, 8 errors, but no faster than Sonnet (mean 5.0s at low,
+    5.8s at medium, max 17-18s). At `high` effort thinking exhausted `max_tokens=4096` and returned an empty
+    response on `loop_experienced_assembly` in 2 of 3 runs.
+  - *Opus 5.5* (thinking cannot be disabled): low 16/16, medium 30/31 reviewed, which is within noise of
+    Sonnet, at 2-2.5x the cost and 6.6s / 9.6s mean. *Fable 5.1*: 15/16 at $0.059/call, with a 37s max that
+    breaks the 30s ceiling.
 - **Prompt caching: evaluated 2026-07 and rejected at current traffic.** ~104 fresh Claude calls/day
   (CloudWatch, 14-day window), only ~35 hours/fortnight above 12 calls/hour, against a 5-minute cache TTL and a
   prefix fragmented by language/arch/audience/type. Generous math: ~$0.40 saved per fortnight of ~$22 spend.
diff --git a/app/model_costs.py b/app/model_costs.py
index e8247d5..0d96282 100644
--- a/app/model_costs.py
+++ b/app/model_costs.py
@@ -19,7 +19,7 @@ class ModelCost(NamedTuple):
 
 
 # Model family costs in USD per million tokens
-# Updated: 2026-09-29 based on https://platform.claude.com/docs/en/about-claude/pricing
+# Updated: 2026-10-10 based on https://platform.claude.com/docs/en/about-claude/pricing
 #
 # Notes:
 # - Opus 4.5+ moved to a new lower price tier ($5/$25) and now bundle the 1M
@@ -32,8 +32,14 @@ class ModelCost(NamedTuple):
 MODEL_FAMILIES = {
     # Claude 5 family. Sonnet 5's launch $2/$10 became its standard price;
     # the planned 2026-09-01 rise to $3/$15 was cancelled.
+    "fable-5.1": ModelCost(10.0, 50.0),
+    "fable-5": ModelCost(10.0, 50.0),
     "sonnet-5.5": ModelCost(2.0, 10.0),
     "sonnet-5": ModelCost(2.0, 10.0),
+    # Haiku 5.5 is the first model with length-tiered pricing: these are the
+    # rates for prompts up to 100k tokens ($0.50/$2.50 above that). Explain
+    # requests are capped far below 100k, so the base tier always applies.
+    "haiku-5.5": ModelCost(0.10, 0.50),
     "opus-5.5": ModelCost(4.0, 20.0),
     "opus-5": ModelCost(5.0, 25.0),
     # Opus 4.5+: new pricing tier, 1M context bundled
diff --git a/app/test_model_costs.py b/app/test_model_costs.py
index 5f41be9..3f15eb2 100644
--- a/app/test_model_costs.py
+++ b/app/test_model_costs.py
@@ -93,6 +93,23 @@ def test_sonnet_5_5_cost(self):
         assert input_cost == 2.0 / 1_000_000
         assert output_cost == 10.0 / 1_000_000
 
+    def test_haiku_5_5_cost(self):
+        """Haiku 5.5 base tier (prompts up to 100k tokens)."""
+        input_cost, output_cost = get_model_cost("claude-haiku-5-5")
+        assert input_cost == 0.10 / 1_000_000
+        assert output_cost == 0.50 / 1_000_000
+
+    def test_opus_5_5_cost(self):
+        input_cost, output_cost = get_model_cost("claude-opus-5-5")
+        assert input_cost == 4.0 / 1_000_000
+        assert output_cost == 20.0 / 1_000_000
+
+    def test_fable_cost(self):
+        for model in ("claude-fable-5", "claude-fable-5-1"):
+            input_cost, output_cost = get_model_cost(model)
+            assert input_cost == 10.0 / 1_000_000
+            assert output_cost == 50.0 / 1_000_000
+
     def test_opus_4_cost(self):
         """Test Claude 4 Opus costs (legacy $15/$75 pricing)."""
         input_cost, output_cost = get_model_cost("claude-opus-4-0")