metergraph-core 0.8.4__tar.gz → 0.8.6__tar.gz

{metergraph_core-0.8.4/src/metergraph_core.egg-info → metergraph_core-0.8.6}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: metergraph-core
-Version: 0.8.4
+Version: 0.8.6
 Summary: Open, effective-dated LLM pricing catalog and deterministic cost calculator for popular models across major providers and gateways, including OpenAI, Anthropic, Google Gemini, AWS Bedrock, OpenRouter, Vercel AI Gateway, and more.
 Author: Metergraph
 License-Expression: Apache-2.0
@@ -603,3 +603,33 @@ lifecycles.
 The model registry version is an ISO date. When more than one registry ships on
 the same date, append a positive revision such as `2026-09-25.2` so consumers
 can distinguish the policies while pricing validation still uses that date.
+
+
+## Complete call pricing
+
+`price_call_from_response` combines token costs with supported retrieval fees.
+`cost_from_response` remains token-only. Retrieval pricing supports Anthropic
+`web_search` and `web_fetch` on `anthropic-api`; other observed scopes are unresolved.
+
+```python
+from metergraph_core import price_call_from_response
+
+result = price_call_from_response(
+    "anthropic",
+    {"type": "message", "model": "claude-sonnet-5-5", "usage": {
+        "input_tokens": 1000, "output_tokens": 100,
+        "server_tool_use": {"web_search_requests": 2},
+    }},
+    at="2026-10-20T00:00:00Z",
+)
+assert str(result.cost_usd) == "0.02300000"
+```
+
+`cost_usd` is set only when the call is fully priced. `known_subtotal_usd`
+contains exact components; `estimated_total_usd` separates token fallback
+estimates when all retrieval fees are known. Missing charges are reported in
+`missing_dimensions`. Counts must be nonnegative integers.
+
+For normalized usage, use `price_call` with `ModelUsage` and `RetrievalUsage`.
+To compose an existing token result from the same catalog, use `compose_call_cost`.
+`resolve_billing` accepts complete-call results for gateway reconciliation.

{metergraph_core-0.8.4 → metergraph_core-0.8.6}/README.md

@@ -573,3 +573,33 @@ lifecycles.
 The model registry version is an ISO date. When more than one registry ships on
 the same date, append a positive revision such as `2026-09-25.2` so consumers
 can distinguish the policies while pricing validation still uses that date.
+
+
+## Complete call pricing
+
+`price_call_from_response` combines token costs with supported retrieval fees.
+`cost_from_response` remains token-only. Retrieval pricing supports Anthropic
+`web_search` and `web_fetch` on `anthropic-api`; other observed scopes are unresolved.
+
+```python
+from metergraph_core import price_call_from_response
+
+result = price_call_from_response(
+    "anthropic",
+    {"type": "message", "model": "claude-sonnet-5-5", "usage": {
+        "input_tokens": 1000, "output_tokens": 100,
+        "server_tool_use": {"web_search_requests": 2},
+    }},
+    at="2026-10-20T00:00:00Z",
+)
+assert str(result.cost_usd) == "0.02300000"
+```
+
+`cost_usd` is set only when the call is fully priced. `known_subtotal_usd`
+contains exact components; `estimated_total_usd` separates token fallback
+estimates when all retrieval fees are known. Missing charges are reported in
+`missing_dimensions`. Counts must be nonnegative integers.
+
+For normalized usage, use `price_call` with `ModelUsage` and `RetrievalUsage`.
+To compose an existing token result from the same catalog, use `compose_call_cost`.
+`resolve_billing` accepts complete-call results for gateway reconciliation.

{metergraph_core-0.8.4 → metergraph_core-0.8.6}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "metergraph-core"
-version = "0.8.4"
+version = "0.8.6"
 description = "Open, effective-dated LLM pricing catalog and deterministic cost calculator for popular models across major providers and gateways, including OpenAI, Anthropic, Google Gemini, AWS Bedrock, OpenRouter, Vercel AI Gateway, and more."
 readme = "README.md"
 license = "Apache-2.0"

{metergraph_core-0.8.4 → metergraph_core-0.8.6}/src/metergraph_core/__init__.py

@@ -155,3 +155,9 @@ __all__ = [
     "search_operation_for_channel",
     "validate_model_registry",
 ]
+
+from .calls import (CallCostResult, RetrievalComponent, RetrievalUsage, compose_call_cost,
+                    price_call, price_call_from_response, response_retrieval_usage)
+
+__all__ += ["CallCostResult", "RetrievalComponent", "RetrievalUsage", "compose_call_cost",
+            "price_call", "price_call_from_response", "response_retrieval_usage"]

{metergraph_core-0.8.4 → metergraph_core-0.8.6}/src/metergraph_core/billing.py

@@ -168,7 +168,7 @@ def _disputed(
 
 
 def resolve_billing(
-    catalog_result: CostResult,
+    catalog_result: "CostResult | CallCostResult",
     evidence: GatewayBillingEvidence,
     *,
     disputes: Iterable[Dispute] = (),
@@ -231,3 +231,7 @@ def resolve_billing(
         catalog_price_id=catalog_result.price_id,
         catalog_reasons=reasons,
     )
+
+from typing import TYPE_CHECKING
+if TYPE_CHECKING:
+    from .calls import CallCostResult

metergraph_core-0.8.6/src/metergraph_core/calls.py

@@ -0,0 +1,188 @@
+"""Compose token and counted retrieval costs from one catalog snapshot."""
+from __future__ import annotations
+
+from dataclasses import dataclass
+from decimal import Decimal
+from typing import Any, Mapping, Sequence
+
+from .catalog import CostResult, ModelUsage, direct_channel_for_provider, normalize_provider
+from .loader import LoadedCatalog
+from .responses import _as_mapping, _bundled_catalog, _first, cost_from_response, response_format
+
+_ZERO = Decimal('0')
+
+
+@dataclass(frozen=True, slots=True)
+class RetrievalUsage:
+    operation: str
+    units: Any
+
+
+@dataclass(frozen=True, slots=True)
+class RetrievalComponent:
+    operation: str
+    units: Any
+    cost_usd: Decimal | None
+    price_id: str | None
+    basis: str
+    reasons: tuple[str, ...] = ()
+    source_url: str | None = None
+
+
+@dataclass(frozen=True, slots=True)
+class CallCostResult:
+    token_cost: CostResult
+    retrieval: tuple[RetrievalComponent, ...]
+    cost_usd: Decimal | None
+    known_subtotal_usd: Decimal
+    estimated_total_usd: Decimal | None
+    status: str
+    token_basis: str
+    missing_dimensions: tuple[str, ...]
+    catalog_hash: str
+
+
+    @property
+    def publisher(self) -> str | None:
+        return self.token_cost.publisher
+
+    @property
+    def price_id(self) -> str | None:
+        return self.token_cost.price_id
+
+    @property
+    def reasons(self) -> tuple[str, ...]:
+        return tuple(dict.fromkeys((*self.token_cost.reasons,
+            *(f"missing_{d}" for d in self.missing_dimensions))))
+
+
+def compose_call_cost(
+    token_cost: CostResult, *, catalog: LoadedCatalog,
+    retrieval: Sequence[RetrievalUsage] = (), provider: str | None = None,
+    channel: str | None = None, region: str = 'global',
+) -> CallCostResult:
+    """Combine retrieval with a token result from the same catalog and time.
+
+    Partial token results remain estimates, excluded from the known subtotal.
+    """
+    if token_cost.effective_at is None:
+        raise ValueError('token_cost must carry its effective_at')
+    if token_cost.price is not None:
+        selected = catalog.lookup_price(model=token_cost.canonical_model,
+            channel=token_cost.price.pricing_channel, region=token_cost.price.region,
+            at=token_cost.effective_at)
+        if selected.price != token_cost.price:
+            raise ValueError('token price does not match the catalog snapshot')
+    if len({item.operation for item in retrieval}) != len(retrieval):
+        raise ValueError('retrieval operations must be aggregated once per call')
+    resolved_channel = token_cost.price.pricing_channel if token_cost.price else (channel or direct_channel_for_provider(provider))
+    publisher = token_cost.publisher or normalize_provider(provider or '')
+    missing = list(token_cost.missing_dimensions)
+    token_basis = ('exact' if token_cost.status == 'priced' else
+                   'fallback_estimate' if token_cost.cost_usd is not None else 'unknown')
+    # Retrieval rates are USD. Do not add them to a different currency.
+    currency_ok = token_cost.currency in (None, 'USD')
+    known = token_cost.cost_usd if token_basis == 'exact' and currency_ok else _ZERO
+    if not currency_ok:
+        missing.append('currency')
+        token_basis = 'unknown'
+    parts = []
+    for usage in retrieval:
+        if resolved_channel == 'anthropic-api' and publisher == 'anthropic' and usage.operation in {'web_search', 'web_fetch'}:
+            fee = catalog.price_retrieval(channel=resolved_channel, operation=usage.operation,
+                units=usage.units, at=token_cost.effective_at, region=region)
+            part = RetrievalComponent(usage.operation, usage.units, fee.cost_usd,
+                fee.price_id, 'exact' if fee.status == 'priced' else 'unknown', fee.reasons, fee.source_url)
+        else:
+            part = RetrievalComponent(usage.operation, usage.units, None, None,
+                'unknown', ('unsupported_retrieval_scope',))
+        part = RetrievalComponent(part.operation,
+            part.units if type(part.units) is int and part.units >= 0 else None,
+            part.cost_usd, part.price_id, part.basis, part.reasons, part.source_url)
+        parts.append(part)
+        if part.cost_usd is None:
+            missing.append(f'retrieval.{usage.operation}')
+        else:
+            known += part.cost_usd
+    unresolved = any(p.basis == 'unknown' for p in parts)
+    status = ('unpriced' if token_cost.price is None or token_cost.status == 'unpriced' else
+              'partial' if token_basis != 'exact' or unresolved or missing else 'priced')
+    estimated = None
+    if token_basis == 'fallback_estimate' and not unresolved:
+        estimated = token_cost.cost_usd + sum((p.cost_usd or _ZERO for p in parts), _ZERO)
+    return CallCostResult(token_cost, tuple(parts), known if status == 'priced' else None,
+        known, estimated, status, token_basis, tuple(dict.fromkeys(missing)), catalog.content_hash)
+
+
+def price_call(
+    *, model: Any, usage: ModelUsage, at: Any, catalog: LoadedCatalog | None = None,
+    provider: str | None = None, channel: str | None = None, region: str | None = None,
+    retrieval: Sequence[RetrievalUsage] = (),
+) -> CallCostResult:
+    """Price normalized token and retrieval usage against one snapshot."""
+    loaded = catalog if catalog is not None else _bundled_catalog()
+    tokens = loaded.calculate_cost(model=model, usage=usage, at=at, provider=provider,
+        channel=channel, region=region)
+    return compose_call_cost(tokens, catalog=loaded, retrieval=retrieval, provider=provider,
+        channel=channel or direct_channel_for_provider(provider), region=region or 'global')
+
+
+def response_retrieval_usage(response: Any, *, format: str | None = None) -> tuple[RetrievalUsage, ...]:
+    """Read operation metadata without reading prompt or response text.
+
+    Tool evidence without a valid usage counter remains unresolved.
+    """
+    data = _as_mapping(response)
+    detected = format or response_format(data)
+    if detected == 'anthropic.messages':
+        usage = data.get('usage')
+        usage = usage if isinstance(usage, Mapping) else {}
+        server = usage.get('server_tool_use')
+        if server is not None and not isinstance(server, Mapping):
+            return (RetrievalUsage('web_search', None),)
+        server = server or {}
+        found = [RetrievalUsage(op, server[key]) for key, op in (
+            ('web_search_requests', 'web_search'), ('web_fetch_requests', 'web_fetch')) if key in server]
+        observed = set()
+        content = data.get('content')
+        if isinstance(content, list):
+            for block in content:
+                if isinstance(block, Mapping) and block.get('type') == 'server_tool_use':
+                    name = block.get('name')
+                    if name in ('web_search', 'web_fetch'):
+                        observed.add(name)
+        found.extend(RetrievalUsage(op, None) for op in sorted(observed) if op not in {f.operation for f in found})
+        return tuple(found)
+    if detected == 'gemini.generate_content':
+        candidates = data.get('candidates')
+        if isinstance(candidates, list):
+            for item in candidates:
+                metadata = _first(item, 'groundingMetadata', 'grounding_metadata') if isinstance(item, Mapping) else None
+                if isinstance(metadata, Mapping) and any(
+                    _first(metadata, camel, snake) is not None
+                    for camel, snake in (("webSearchQueries", "web_search_queries"),
+                                         ("groundingChunks", "grounding_chunks"))
+                ):
+                    # Grounding evidence does not establish billable query count.
+                    return (RetrievalUsage('google_search_grounding', None),)
+    if detected == 'openai.responses':
+        output = data.get('output')
+        if isinstance(output, list):
+            count = sum(1 for item in output if isinstance(item, Mapping) and item.get('type') == 'web_search_call')
+            if count:
+                return (RetrievalUsage('web_search', count),)
+    return ()
+
+
+def price_call_from_response(
+    provider: str, response: Any, *, catalog: LoadedCatalog | None = None,
+    model: str | None = None, at: Any = None, channel: str | None = None,
+    region: str | None = None, service_tier: str | None = None, format: str | None = None,
+) -> CallCostResult:
+    """Price a provider response, including supported counted retrieval fees."""
+    loaded = catalog if catalog is not None else _bundled_catalog()
+    tokens = cost_from_response(provider, response, catalog=loaded, model=model, at=at,
+        channel=channel, region=region, service_tier=service_tier, format=format)
+    return compose_call_cost(tokens, catalog=loaded,
+        retrieval=response_retrieval_usage(response, format=format), provider=provider,
+        channel=channel or direct_channel_for_provider(provider), region=region or 'global')

{metergraph_core-0.8.4 → metergraph_core-0.8.6}/src/metergraph_core/data/models.yaml

@@ -1,4 +1,4 @@
-version: '2026-10-27'
+version: '2026-10-29'
 models:
 - canonical_id: openai/gpt-5.6-sol
   display_name: GPT-5.6 Sol
@@ -888,6 +888,30 @@ models:
     execution_profiles:
     - default
     display_name: Kimi K3.1 Pro Highspeed (direct)
+- canonical_id: minimax/minimax-m3.1-flash
+  display_name: MiniMax M3.1 Flash
+  publisher: minimax
+  released: '2026-10-19'  # https://www.minimax.io/blog/minimax-m31f
+  routes:
+  - key: default:minimax/minimax-m3.1-flash
+    id: minimax/minimax-m3.1-flash
+    provider: vercel
+    model_id: minimax/minimax-m3.1-flash
+    pricing_channel: vercel-ai-gateway
+    execution_profiles:
+    - default
+- canonical_id: mistral/mistral-small-4
+  display_name: Mistral Small 4
+  publisher: mistral
+  released: '2026-03-16'  # https://docs.mistral.ai/models/mistral-small-4-0-26-03
+  routes:
+  - key: default:mistral:mistral-small-2603
+    id: mistral:mistral-small-2603
+    provider: mistral
+    model_id: mistral-small-2603
+    pricing_channel: mistral-api
+    execution_profiles:
+    - default
 execution_profiles:
 - id: default
   routes:
@@ -963,6 +987,8 @@ execution_profiles:
   - default:openai:gpt-6.1-terra
   - default:moonshotai:kimi-k3.1-pro
   - default:moonshotai:kimi-k3.1-pro-highspeed
+  - default:minimax/minimax-m3.1-flash
+  - default:mistral:mistral-small-2603
 - id: bedrock
   routes:
   - bedrock:anthropic/claude-sonnet-5
@@ -1029,6 +1055,7 @@ offer_groups:
   - default:deepseek/deepseek-v4.1-flash-1017
   - default:deepseek/deepseek-v4.1-flash-1017-fast
   - default:inception/mercury-3
+  - default:minimax/minimax-m3.1-flash
 - id: fireworks
   credential: FIREWORKS_API_KEY
   routes:
@@ -1059,6 +1086,10 @@ offer_groups:
   routes:
   - default:moonshotai:kimi-k3.1-pro
   - default:moonshotai:kimi-k3.1-pro-highspeed
+- id: mistral-direct
+  credential: MISTRAL_API_KEY
+  routes:
+  - default:mistral:mistral-small-2603
 - id: bedrock
   credential: null
   routes:

{metergraph_core-0.8.4 → metergraph_core-0.8.6}/src/metergraph_core/data/prices.yaml

@@ -6,9 +6,9 @@
 #
 # Community updates welcome: add an alias or price entry with its provider
 # source_url and open a PR. CI validates structure and date overlaps.
-version: "2026-10-29"
+version: "2026-10-29.3"
 currency: USD
-pricing_verified_at: "2026-10-27"
+pricing_verified_at: "2026-10-29"
 models:
   - canonical_id: openai/gpt-6-sol
     publisher: openai
@@ -1579,6 +1579,7 @@ models:
       - channel: anthropic-api
         region: global
         effective_from: "2026-10-07"
+        effective_to: "2026-10-29"
         input_per_mtok: 2.00
         output_per_mtok: 10.00
         cache_read_per_mtok: 0.10
@@ -1587,6 +1588,17 @@ models:
         batch_input_per_mtok: 1.00
         batch_output_per_mtok: 5.00
         source_url: https://www.anthropic.com/claude-haiku-5-5
+      - channel: anthropic-api
+        region: global
+        effective_from: "2026-10-29"
+        input_per_mtok: 2.00
+        output_per_mtok: 10.00
+        cache_read_per_mtok: 0.10
+        cache_write_5m_per_mtok: 2.50
+        cache_write_1h_per_mtok: 4.00
+        batch_input_per_mtok: 1.00
+        batch_output_per_mtok: 5.00
+        source_url: https://platform.claude.com/docs/en/about-claude/pricing
       - channel: vercel-ai-gateway
         region: global
         effective_from: "2026-10-07"
@@ -2436,12 +2448,24 @@ models:
       - channel: mistral-api
         region: global
         effective_from: "2026-10-20"
+        effective_to: "2026-10-29"
         input_per_mtok: 1.36
         output_per_mtok: 4.18
         cache_read_per_mtok: 0.14
         batch_input_per_mtok: 0.68
         batch_output_per_mtok: 2.09
         source_url: https://docs.mistral.ai/models/mistral-large
+      # Batch input/output use the documented 50% discount on all models:
+      # https://docs.mistral.ai/studio/batch-processing
+      - channel: mistral-api
+        region: global
+        effective_from: "2026-10-29"
+        input_per_mtok: 0.70
+        output_per_mtok: 2.10
+        cache_read_per_mtok: 0.07
+        batch_input_per_mtok: 0.35
+        batch_output_per_mtok: 1.05
+        source_url: https://docs.mistral.ai/models/mistral-large
       - channel: vercel-ai-gateway
         region: global
         effective_from: "2026-10-06"
@@ -2915,6 +2939,9 @@ models:
         rules:
           uncaptured_fees: true
         source_url: https://docs.perplexity.ai/docs/getting-started/pricing
+      # The Agent API's model list is a separate product: its Sonar rates,
+      # including its cache read rate, do not price this channel, so the
+      # direct token rates above stay the last known ones.
       - channel: vercel-ai-gateway
         region: global
         effective_from: "2026-10-11"
@@ -3661,10 +3688,22 @@ models:
       - channel: moonshot-api
         region: global
         effective_from: "2026-07-16"
+        effective_to: "2026-10-29"
         input_per_mtok: 3.0
         output_per_mtok: 15.0
         cache_read_per_mtok: 0.3
         source_url: https://platform.kimi.ai/docs/pricing/chat
+      # The platform's cache-write figure is stored for the default 5m TTL;
+      # the separately billed 1h tier remains unknown.
+      - channel: moonshot-api
+        region: global
+        effective_from: "2026-10-29"
+        input_per_mtok: 3.0
+        output_per_mtok: 15.0
+        cache_read_per_mtok: 0.3
+        cache_write_5m_per_mtok: 3.0
+        rules: {input_includes_cache_read: true, input_includes_cache_write: true}
+        source_url: https://platform.kimi.ai/
       - channel: fireworks-api
         region: global
         effective_from: "2026-07-19"
@@ -4574,6 +4613,22 @@ models:
       - channel: google-api
         region: global
         effective_from: "2026-08-27"
+        effective_to: "2026-10-28"
+        input_per_mtok: 1.50
+        output_per_mtok: 17.50
+        source_url: https://ai.google.dev/gemini-api/docs/pricing
+      - channel: google-api
+        region: global
+        effective_from: "2026-10-28"
+        effective_to: "2026-10-29"
+        # Reconcile the already-published live interval; this was an automated
+        # text-rate misclassification, corrected by the successor below.
+        input_per_mtok: 1.5
+        output_per_mtok: 9
+        source_url: https://ai.google.dev/gemini-api/docs/pricing
+      - channel: google-api
+        region: global
+        effective_from: "2026-10-29"
         input_per_mtok: 1.50
         output_per_mtok: 17.50
         source_url: https://ai.google.dev/gemini-api/docs/pricing
@@ -5085,6 +5140,7 @@ models:
       - channel: anthropic-api
         region: global
         effective_from: "2026-06-09"
+        effective_to: "2026-10-29"
         input_per_mtok: 10
         output_per_mtok: 50
         cache_read_per_mtok: 1
@@ -5093,6 +5149,17 @@ models:
         batch_input_per_mtok: 5
         batch_output_per_mtok: 25
         source_url: https://ai-gateway.vercel.sh/v1/models
+      - channel: anthropic-api
+        region: global
+        effective_from: "2026-10-29"
+        input_per_mtok: 10
+        output_per_mtok: 50
+        cache_read_per_mtok: 1
+        cache_write_5m_per_mtok: 12.5
+        cache_write_1h_per_mtok: 20
+        batch_input_per_mtok: 5
+        batch_output_per_mtok: 25
+        source_url: https://platform.claude.com/docs/en/about-claude/pricing
   - canonical_id: anthropic/claude-opus-4
     publisher: anthropic
     aliases:
@@ -5113,6 +5180,7 @@ models:
       - channel: anthropic-api
         region: global
         effective_from: "2025-05-14"
+        effective_to: "2026-10-29"
         input_per_mtok: 15
         output_per_mtok: 75
         cache_read_per_mtok: 1.5
@@ -5121,6 +5189,17 @@ models:
         batch_input_per_mtok: 7.5
         batch_output_per_mtok: 37.5
         source_url: https://ai-gateway.vercel.sh/v1/models
+      - channel: anthropic-api
+        region: global
+        effective_from: "2026-10-29"
+        input_per_mtok: 15
+        output_per_mtok: 75
+        cache_read_per_mtok: 1.5
+        cache_write_5m_per_mtok: 18.75
+        cache_write_1h_per_mtok: 30
+        batch_input_per_mtok: 7.5
+        batch_output_per_mtok: 37.5
+        source_url: https://platform.claude.com/docs/en/about-claude/pricing
       - channel: vercel-ai-gateway
         region: global
         effective_from: "2026-10-02"
@@ -5148,6 +5227,7 @@ models:
       - channel: anthropic-api
         region: global
         effective_from: "2025-08-05"
+        effective_to: "2026-10-29"
         input_per_mtok: 15
         output_per_mtok: 75
         cache_read_per_mtok: 1.5
@@ -5156,6 +5236,17 @@ models:
         batch_input_per_mtok: 7.5
         batch_output_per_mtok: 37.5
         source_url: https://ai-gateway.vercel.sh/v1/models
+      - channel: anthropic-api
+        region: global
+        effective_from: "2026-10-29"
+        input_per_mtok: 15
+        output_per_mtok: 75
+        cache_read_per_mtok: 1.5
+        cache_write_5m_per_mtok: 18.75
+        cache_write_1h_per_mtok: 30
+        batch_input_per_mtok: 7.5
+        batch_output_per_mtok: 37.5
+        source_url: https://platform.claude.com/docs/en/about-claude/pricing
   - canonical_id: anthropic/claude-opus-4.5
     publisher: anthropic
     aliases:
@@ -5166,6 +5257,7 @@ models:
       - channel: anthropic-api
         region: global
         effective_from: "2025-11-01"
+        effective_to: "2026-10-29"
         input_per_mtok: 5
         output_per_mtok: 25
         cache_read_per_mtok: 0.5
@@ -5174,6 +5266,17 @@ models:
         batch_input_per_mtok: 2.5
         batch_output_per_mtok: 12.5
         source_url: https://ai-gateway.vercel.sh/v1/models
+      - channel: anthropic-api
+        region: global
+        effective_from: "2026-10-29"
+        input_per_mtok: 5
+        output_per_mtok: 25
+        cache_read_per_mtok: 0.5
+        cache_write_5m_per_mtok: 6.25
+        cache_write_1h_per_mtok: 10
+        batch_input_per_mtok: 2.5
+        batch_output_per_mtok: 12.5
+        source_url: https://platform.claude.com/docs/en/about-claude/pricing
       - channel: vercel-ai-gateway
         region: global
         effective_from: "2025-11-01"
@@ -5236,6 +5339,7 @@ models:
       - channel: anthropic-api
         region: global
         effective_from: "2025-05-14"
+        effective_to: "2026-10-29"
         input_per_mtok: 3
         output_per_mtok: 15
         cache_read_per_mtok: 0.3
@@ -5244,6 +5348,17 @@ models:
         batch_input_per_mtok: 1.5
         batch_output_per_mtok: 7.5
         source_url: https://ai-gateway.vercel.sh/v1/models
+      - channel: anthropic-api
+        region: global
+        effective_from: "2026-10-29"
+        input_per_mtok: 3
+        output_per_mtok: 15
+        cache_read_per_mtok: 0.3
+        cache_write_5m_per_mtok: 3.75
+        cache_write_1h_per_mtok: 6
+        batch_input_per_mtok: 1.5
+        batch_output_per_mtok: 7.5
+        source_url: https://platform.claude.com/docs/en/about-claude/pricing
       - channel: vercel-ai-gateway
         region: global
         effective_from: "2025-05-14"
@@ -5416,6 +5531,9 @@ models:
         input_per_mtok: 0.15
         output_per_mtok: 0.6
         cache_read_per_mtok: 0.015
+        # prompt_tokens includes cached_tokens: https://docs.mistral.ai/studio/conversations/advanced/prompt-caching
+        rules:
+          input_includes_cache_read: true
         source_url: https://docs.mistral.ai/inference/pricing
       - channel: openrouter-api
         region: global
@@ -5743,12 +5861,22 @@ models:
       - channel: xai-api
         region: global
         effective_from: "2026-03-10"
+        effective_to: "2026-10-29"
         input_per_mtok: 1.25
         output_per_mtok: 2.5
         cache_read_per_mtok: 0.2
         rules:
           long_context: {threshold: 200000, input_multiplier: 2.0, output_multiplier: 2.0}
         source_url: https://ai-gateway.vercel.sh/v1/models
+      - channel: xai-api
+        region: global
+        effective_from: "2026-10-29"
+        input_per_mtok: 1.25
+        output_per_mtok: 2.5
+        cache_read_per_mtok: 0.2
+        rules:
+          long_context: {threshold: 200000, input_multiplier: 2.0, output_multiplier: 2.0}
+        source_url: https://docs.x.ai/developers/models
   - canonical_id: xai/grok-4.20-reasoning
     publisher: xai
     aliases:
@@ -5781,12 +5909,22 @@ models:
       - channel: xai-api
         region: global
         effective_from: "2026-03-10"
+        effective_to: "2026-10-29"
         input_per_mtok: 1.25
         output_per_mtok: 2.5
         cache_read_per_mtok: 0.2
         rules:
           long_context: {threshold: 200000, input_multiplier: 2.0, output_multiplier: 2.0}
         source_url: https://ai-gateway.vercel.sh/v1/models
+      - channel: xai-api
+        region: global
+        effective_from: "2026-10-29"
+        input_per_mtok: 1.25
+        output_per_mtok: 2.5
+        cache_read_per_mtok: 0.2
+        rules:
+          long_context: {threshold: 200000, input_multiplier: 2.0, output_multiplier: 2.0}
+        source_url: https://docs.x.ai/developers/models
       - channel: openrouter-api
         region: global
         effective_from: "2026-10-29"
@@ -6147,6 +6285,10 @@ models:
         alias: minimax/minimax-m3.1-flash
         channel: vercel-ai-gateway
         rules: {input_includes_cache_read: true}
+      - provider: vercel
+        alias: minimax/minimax-m3.1-flash
+        channel: vercel-ai-gateway
+        rules: {input_includes_cache_read: true}
     prices:
       - channel: minimax-api
         region: global
@@ -6321,6 +6463,14 @@ models:
           input_includes_cache_write: true
         effective_from: "2026-11-02"
         source_url: https://openrouter.ai/qwen/qwen3-coder-30b-a3b-instruct
+      - provider: openrouter
+        alias: alibaba/qwen3-coder-30b-a3b-instruct
+        channel: openrouter-api
+        effective_from: "2026-11-02"
+        rules:
+          input_includes_cache_read: true
+          input_includes_cache_write: true
+        source_url: https://openrouter.ai/qwen/qwen3-coder-30b-a3b-instruct
     prices:
       - channel: vercel-ai-gateway
         region: global
@@ -12111,11 +12261,22 @@ models:
       - channel: openai-api
         region: global
         effective_from: "2025-06-10"
+        effective_to: "2026-10-29"
+        input_per_mtok: 2
+        output_per_mtok: 8
+        cache_read_per_mtok: 0.50
+        batch_input_per_mtok: 1
+        batch_output_per_mtok: 4
+        source_url: https://developers.openai.com/api/docs/models/o3
+      - channel: openai-api
+        region: global
+        effective_from: "2026-10-29"
         input_per_mtok: 2
         output_per_mtok: 8
         cache_read_per_mtok: 0.50
         batch_input_per_mtok: 1
         batch_output_per_mtok: 4
+        batch_cache_read_per_mtok: 0.25
         source_url: https://developers.openai.com/api/docs/models/o3
       - channel: vercel-ai-gateway
         region: global

{metergraph_core-0.8.4 → metergraph_core-0.8.6}/src/metergraph_core/extractors.py

@@ -215,7 +215,7 @@ def parse_google(html_text: str, *, duplicates: list[str] | None = None) -> Tabl
     the one in effect; per-image output is not a token rate. A label may name
     the media it covers, as in "Input price (text, image, video)"."""
     parts = page_text(html_text).split("Try it in Google AI Studio")
-    tier = r"(?:Free of charge \| |Not available \| )?([^|]*\$[^|]*)"
+    tier = r"(?:Free of charge \| |Not available \| )?([^|]*\$[^|]*(?:\| \$[^|]*)*)"
     table: Table = {}
     for before, block in zip(parts, parts[1:]):
         ids = _google_ids([c.strip() for c in before.split("|") if c.strip()])
@@ -226,7 +226,12 @@ def parse_google(html_text: str, *, duplicates: list[str] | None = None) -> Tabl
         if not (inp and out):
             continue
         cache = re.search(r"Context caching price[^|]*\| " + tier, block)
-        output = None if "per image" in out.group(1) else dollars(out.group(1))
+        # A scalar output rate cannot express separate text/video/audio prices.
+        # Leave it missing so refresh escalates rather than selecting the first
+        # modality and overwriting a manually verified catalog rate.
+        output_cell = out.group(1)
+        modalities = set(re.findall(r"\((text|video|audio)\)", output_cell, re.I))
+        output = None if "per image" in output_cell or len(modalities) > 1 else dollars(output_cell)
         rates = {"input": dollars(inp.group(1)), "output": output, "cache_read": dollars(cache.group(1)) if cache else None}
         for model_id in ids:
             _keep_first(table, model_id, {k: v for k, v in rates.items() if v is not None}, duplicates)

{metergraph_core-0.8.4 → metergraph_core-0.8.6}/src/metergraph_core/loader.py

@@ -89,16 +89,25 @@ class LoadedCatalog:
     # Gateways whose amount billing declines per publisher; see Dispute.
     disputes: tuple[Dispute, ...] = ()
 
-    def canonical_model_id(self, provider: str, model_id: str) -> str:
-        """Resolve a captured provider-native model id to its canonical id,
-        using the known source ``provider`` to disambiguate names shared across
-        providers. An unknown provider, unknown id, or non-string input is left
-        unchanged so none is ever misattributed."""
+    def canonical_model_id(self, provider: str, model_id: str, *, at: Any = None) -> str:
+        """Resolve a provider's model name, using ``at`` (default now) for migrations.
+
+        Unknown or ambiguous names are returned unchanged.
+        """
         if not isinstance(model_id, str) or not isinstance(provider, str):
             return model_id
-        return self.canonical_ids.get(
-            (normalize_provider(provider), model_id), model_id
-        )
+        provider = normalize_provider(provider)
+        canonical = self.canonical_ids.get((provider, model_id))
+        if canonical is not None:
+            return canonical
+        when = _coerce_datetime(at) if at is not None else datetime.now(timezone.utc)
+        matches = {
+            alias.canonical_id
+            for alias in self.snapshot._aliases.get((provider, model_id.lower()), ())
+            if (alias.effective_from is None or alias.effective_from <= when)
+            and (alias.effective_to is None or when < alias.effective_to)
+        }
+        return next(iter(matches)) if len(matches) == 1 else model_id
 
     def infer_direct_channel(self, model: Any) -> str | None:
         """Infer a direct pricing channel only when the catalog is unambiguous."""

{metergraph_core-0.8.4 → metergraph_core-0.8.6}/src/metergraph_core/models.py

@@ -19,13 +19,14 @@ EXECUTION_PROFILES = frozenset({"default", "bedrock"})
 # "moonshotai", the name its catalog aliases carry ("moonshot" is only an
 # input alias the catalog normalizes).
 ROUTE_PROVIDERS = frozenset(
-    {"anthropic", "bedrock", "fireworks", "moonshotai", "openai", "vercel"}
+    {"anthropic", "bedrock", "fireworks", "mistral", "moonshotai", "openai", "vercel"}
 )
 ROUTE_PROVIDER_CHANNELS = MappingProxyType(
     {
         "anthropic": "anthropic-api",
         "bedrock": "aws-bedrock",
         "fireworks": "fireworks-api",
+        "mistral": "mistral-api",
         "moonshotai": "moonshot-api",
         "openai": "openai-api",
         "vercel": "vercel-ai-gateway",
@@ -38,6 +39,7 @@ OFFER_GROUP_PROFILES = MappingProxyType(
         "openai-direct": "default",
         "anthropic-direct": "default",
         "moonshot-direct": "default",
+        "mistral-direct": "default",
         "bedrock": "bedrock",
     }
 )
@@ -48,6 +50,7 @@ OFFER_GROUP_CREDENTIALS = MappingProxyType(
         "openai-direct": "OPENAI_API_KEY",
         "anthropic-direct": "ANTHROPIC_API_KEY",
         "moonshot-direct": "MOONSHOT_API_KEY",
+        "mistral-direct": "MISTRAL_API_KEY",
         "bedrock": None,
     }
 )
@@ -58,6 +61,7 @@ OFFER_GROUP_PROVIDERS = MappingProxyType(
         "openai-direct": "openai",
         "anthropic-direct": "anthropic",
         "moonshot-direct": "moonshotai",
+        "mistral-direct": "mistral",
         "bedrock": "bedrock",
     }
 )
@@ -475,7 +479,7 @@ def validate_model_registry(
                 f"{route.id}: model {route.model_id!r} has no pricing channel "
                 f"{route.pricing_channel!r}"
             )
-        canonical = catalog.canonical_model_id(route.provider, route.model_id)
+        canonical = catalog.canonical_model_id(route.provider, route.model_id, at=validation_time)
         if canonical == route.model_id:
             resolved = catalog.snapshot.resolve_price(
                 model=route.model_id,

{metergraph_core-0.8.4 → metergraph_core-0.8.6}/src/metergraph_core/retrieval.py

@@ -52,6 +52,7 @@ class RetrievalCostResult:
     cost_usd: Decimal | None
     status: str
     reasons: tuple[str, ...] = ()
+    source_url: str | None = None
 
 
 # The retrieval operation a counted search or fetch is billed as on each
@@ -231,4 +232,4 @@ class RetrievalCatalog:
         cost = (Decimal(count) * price.per_1k_usd / _THOUSAND).quantize(
             _COST_QUANTUM, rounding=ROUND_HALF_UP
         )
-        return RetrievalCostResult(price.id, cost, "priced", ())
+        return RetrievalCostResult(price.id, cost, "priced", (), price.source_url)

{metergraph_core-0.8.4 → metergraph_core-0.8.6/src/metergraph_core.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: metergraph-core
-Version: 0.8.4
+Version: 0.8.6
 Summary: Open, effective-dated LLM pricing catalog and deterministic cost calculator for popular models across major providers and gateways, including OpenAI, Anthropic, Google Gemini, AWS Bedrock, OpenRouter, Vercel AI Gateway, and more.
 Author: Metergraph
 License-Expression: Apache-2.0
@@ -603,3 +603,33 @@ lifecycles.
 The model registry version is an ISO date. When more than one registry ships on
 the same date, append a positive revision such as `2026-09-25.2` so consumers
 can distinguish the policies while pricing validation still uses that date.
+
+
+## Complete call pricing
+
+`price_call_from_response` combines token costs with supported retrieval fees.
+`cost_from_response` remains token-only. Retrieval pricing supports Anthropic
+`web_search` and `web_fetch` on `anthropic-api`; other observed scopes are unresolved.
+
+```python
+from metergraph_core import price_call_from_response
+
+result = price_call_from_response(
+    "anthropic",
+    {"type": "message", "model": "claude-sonnet-5-5", "usage": {
+        "input_tokens": 1000, "output_tokens": 100,
+        "server_tool_use": {"web_search_requests": 2},
+    }},
+    at="2026-10-20T00:00:00Z",
+)
+assert str(result.cost_usd) == "0.02300000"
+```
+
+`cost_usd` is set only when the call is fully priced. `known_subtotal_usd`
+contains exact components; `estimated_total_usd` separates token fallback
+estimates when all retrieval fees are known. Missing charges are reported in
+`missing_dimensions`. Counts must be nonnegative integers.
+
+For normalized usage, use `price_call` with `ModelUsage` and `RetrievalUsage`.
+To compose an existing token result from the same catalog, use `compose_call_cost`.
+`resolve_billing` accepts complete-call results for gateway reconciliation.

{metergraph_core-0.8.4 → metergraph_core-0.8.6}/src/metergraph_core.egg-info/SOURCES.txt

@@ -3,6 +3,7 @@ README.md
 pyproject.toml
 src/metergraph_core/__init__.py
 src/metergraph_core/billing.py
+src/metergraph_core/calls.py
 src/metergraph_core/catalog.py
 src/metergraph_core/catalog_delta.py
 src/metergraph_core/extractors.py