fix(bedrock): price Claude prompt-cache tokens in /usage (#50307)

Bedrock Claude routes through the AnthropicBedrock SDK and injects
cache_control, so cached tokens are always reported — but the pricing
table had no cache cost fields for any Bedrock model, so /usage showed
"cost unknown" on every cached session. Also, cross-region inference
profiles (us./global./eu. prefixes) never matched the bare pricing keys.

- Add cache_read/cache_write rates to the four Bedrock Claude rows
  (read 0.1x input, write 1.25x input per the Bedrock pricing page).
- Normalize the cross-region prefix in the Bedrock pricing lookup,
  mirroring is_anthropic_bedrock_model's prefix list.

Closes #50295.

agent/usage_pricing.py

@@ -451,6 +451,8 @@ _OFFICIAL_DOCS_PRICING: Dict[tuple[str, str], PricingEntry] = {
     ): PricingEntry(
         input_cost_per_million=Decimal("15.00"),
         output_cost_per_million=Decimal("75.00"),
+        cache_read_cost_per_million=Decimal("1.50"),
+        cache_write_cost_per_million=Decimal("18.75"),
         source="official_docs_snapshot",
         source_url="https://aws.amazon.com/bedrock/pricing/",
         pricing_version="bedrock-pricing-2026-04",
@@ -461,6 +463,8 @@ _OFFICIAL_DOCS_PRICING: Dict[tuple[str, str], PricingEntry] = {
     ): PricingEntry(
         input_cost_per_million=Decimal("3.00"),
         output_cost_per_million=Decimal("15.00"),
+        cache_read_cost_per_million=Decimal("0.30"),
+        cache_write_cost_per_million=Decimal("3.75"),
         source="official_docs_snapshot",
         source_url="https://aws.amazon.com/bedrock/pricing/",
         pricing_version="bedrock-pricing-2026-04",
@@ -471,6 +475,8 @@ _OFFICIAL_DOCS_PRICING: Dict[tuple[str, str], PricingEntry] = {
     ): PricingEntry(
         input_cost_per_million=Decimal("3.00"),
         output_cost_per_million=Decimal("15.00"),
+        cache_read_cost_per_million=Decimal("0.30"),
+        cache_write_cost_per_million=Decimal("3.75"),
         source="official_docs_snapshot",
         source_url="https://aws.amazon.com/bedrock/pricing/",
         pricing_version="bedrock-pricing-2026-04",
@@ -481,6 +487,8 @@ _OFFICIAL_DOCS_PRICING: Dict[tuple[str, str], PricingEntry] = {
     ): PricingEntry(
         input_cost_per_million=Decimal("0.80"),
         output_cost_per_million=Decimal("4.00"),
+        cache_read_cost_per_million=Decimal("0.08"),
+        cache_write_cost_per_million=Decimal("1.00"),
         source="official_docs_snapshot",
         source_url="https://aws.amazon.com/bedrock/pricing/",
         pricing_version="bedrock-pricing-2026-04",
@@ -584,6 +592,26 @@ def resolve_billing_route(
     return BillingRoute(provider=provider_name or "unknown", model=model.split("/")[-1] if model else "", base_url=base_url or "", billing_mode="unknown")
 
 
+def _normalize_bedrock_model_name(model: str) -> str:
+    """Normalize a Bedrock model id to its bare foundation-model form.
+
+    Bedrock cross-region inference profiles prefix the foundation model id
+    with a region scope (``us.`` / ``global.`` / ``eu.`` / ``ap.`` / ``jp.``),
+    e.g. ``us.anthropic.claude-opus-4-7``.  The pricing table is keyed on the
+    bare ``anthropic.claude-*`` id, so the prefix must be stripped before the
+    lookup or every cross-region session prices as unknown.  Mirrors the
+    prefix list in ``bedrock_adapter.is_anthropic_bedrock_model``.  Also
+    normalizes dot-notation version numbers (``4.7`` → ``4-7``).
+    """
+    name = model.lower().strip()
+    for prefix in ("us.", "global.", "eu.", "ap.", "jp."):
+        if name.startswith(prefix):
+            name = name[len(prefix):]
+            break
+    name = re.sub(r"(\d+)\.(\d+)", r"\1-\2", name)
+    return name
+
+
 def _normalize_anthropic_model_name(model: str) -> str:
     """Normalize Anthropic model name variants to canonical form.
 
@@ -614,6 +642,14 @@ def _lookup_official_docs_pricing(route: BillingRoute) -> Optional[PricingEntry]
             entry = _OFFICIAL_DOCS_PRICING.get((route.provider, normalized))
             if entry:
                 return entry
+    # Bedrock cross-region inference profiles carry a region prefix
+    # (us./global./eu./...) that the bare pricing keys don't have.
+    if route.provider == "bedrock":
+        normalized = _normalize_bedrock_model_name(model)
+        if normalized != model:
+            entry = _OFFICIAL_DOCS_PRICING.get((route.provider, normalized))
+            if entry:
+                return entry
     return None
 
 

tests/agent/test_usage_pricing.py

@@ -250,3 +250,75 @@ def test_deepseek_v4_pro_estimate_usage_cost():
     assert result.amount_usd is not None
     # 1M input × $1.74/M + 500K output × $3.48/M = $1.74 + $1.74 = $3.48
     assert float(result.amount_usd) == 3.48
+
+
+def test_bedrock_claude_rows_all_carry_cache_pricing():
+    """Invariant: every Bedrock Claude pricing row must carry cache-read AND
+    cache-write rates, otherwise a cached session prices as ``unknown``.
+
+    Bedrock Claude routes through the AnthropicBedrock SDK and injects
+    cache_control, so cached tokens are always reported — the pricing layer
+    must be able to value them.  See #50295.
+    """
+    from agent.usage_pricing import _OFFICIAL_DOCS_PRICING
+
+    claude_rows = [
+        (prov, model)
+        for (prov, model) in _OFFICIAL_DOCS_PRICING
+        if prov == "bedrock" and "claude" in model
+    ]
+    assert claude_rows, "expected at least one bedrock Claude pricing row"
+    for key in claude_rows:
+        entry = _OFFICIAL_DOCS_PRICING[key]
+        assert entry.input_cost_per_million is not None, key
+        assert entry.cache_read_cost_per_million is not None, key
+        assert entry.cache_write_cost_per_million is not None, key
+        # Cache reads are cheaper than fresh input; cache writes cost more.
+        assert entry.cache_read_cost_per_million < entry.input_cost_per_million, key
+        assert entry.cache_write_cost_per_million > entry.input_cost_per_million, key
+
+
+def test_bedrock_cross_region_profile_prefix_resolves_to_pricing():
+    """Cross-region inference profiles (us./global./eu. prefixes) must resolve
+    to the same pricing entry as the bare foundation-model id.  Without prefix
+    normalization, ``us.anthropic.claude-*`` sessions price as unknown.
+    """
+    bedrock_url = "https://bedrock-runtime.us-east-1.amazonaws.com"
+    bare = get_pricing_entry(
+        "anthropic.claude-sonnet-4-5", provider="bedrock", base_url=bedrock_url
+    )
+    assert bare is not None
+    for prefix in ("us.", "global.", "eu."):
+        scoped = get_pricing_entry(
+            f"{prefix}anthropic.claude-sonnet-4-5",
+            provider="bedrock",
+            base_url=bedrock_url,
+        )
+        assert scoped is not None, prefix
+        assert scoped.input_cost_per_million == bare.input_cost_per_million
+        assert scoped.cache_read_cost_per_million == bare.cache_read_cost_per_million
+
+
+def test_bedrock_claude_cached_session_estimates_cost_not_unknown():
+    """A Bedrock Claude session with cache hits must produce a dollar estimate,
+    not ``unknown`` — the user-visible symptom in #50295.
+    """
+    bedrock_url = "https://bedrock-runtime.us-east-1.amazonaws.com"
+    usage = SimpleNamespace(
+        input_tokens=55,
+        output_tokens=7113,
+        cache_read_input_tokens=1369379,
+        cache_creation_input_tokens=42135,
+    )
+    canonical = normalize_usage(usage, provider="bedrock", api_mode="anthropic_messages")
+    assert canonical.cache_read_tokens == 1369379
+    assert canonical.cache_write_tokens == 42135
+
+    result = estimate_usage_cost(
+        "us.anthropic.claude-opus-4-6",
+        canonical,
+        provider="bedrock",
+        base_url=bedrock_url,
+    )
+    assert result.status == "estimated"
+    assert result.amount_usd is not None