metergraph-core 0.7.4__tar.gz → 0.7.6__tar.gz

{metergraph_core-0.7.4/src/metergraph_core.egg-info → metergraph_core-0.7.6}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: metergraph-core
-Version: 0.7.4
+Version: 0.7.6
 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
@@ -186,6 +186,8 @@ standard `cache_read_per_mtok`, as before. A standard call always uses the
 standard cache-read rate. A row with only a batch cache rate fully prices a
 cached batch call, while a cached standard call on it is `partial` with
 `cache_read_rate_unavailable`.
+Batch cache writes use `Price.batch_cache_write_5m_per_mtok` when stated;
+otherwise they retain the row's standard cache-write rate.
 
 A `ModelUsage` count left `None` is unknown, not zero. A missing input or
 output count keeps the subtotal that is known, marks the result `partial`, and

{metergraph_core-0.7.4 → metergraph_core-0.7.6}/README.md

@@ -156,6 +156,8 @@ standard `cache_read_per_mtok`, as before. A standard call always uses the
 standard cache-read rate. A row with only a batch cache rate fully prices a
 cached batch call, while a cached standard call on it is `partial` with
 `cache_read_rate_unavailable`.
+Batch cache writes use `Price.batch_cache_write_5m_per_mtok` when stated;
+otherwise they retain the row's standard cache-write rate.
 
 A `ModelUsage` count left `None` is unknown, not zero. A missing input or
 output count keeps the subtotal that is known, marks the result `partial`, and

{metergraph_core-0.7.4 → metergraph_core-0.7.6}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "metergraph-core"
-version = "0.7.4"
+version = "0.7.6"
 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-0.7.4 → metergraph_core-0.7.6}/src/metergraph_core/catalog.py

@@ -229,18 +229,20 @@ class Price:
     # one. Absent, batch cache reads keep the standard cache-read rate; it is
     # never inferred from another rate.
     batch_cache_read_per_mtok: Decimal | None = None
+    # OpenAI publishes a separate Batch/Flex cache-write rate.
+    batch_cache_write_5m_per_mtok: Decimal | None = None
 
     def __post_init__(self) -> None:
         # The loader checks the rate it reads, but a row built or replaced
         # directly must not bill a negative or undefined cache component.
-        rate = self.batch_cache_read_per_mtok
-        if rate is not None and not (
-            isinstance(rate, Decimal) and rate.is_finite() and rate >= 0
-        ):
-            raise ValueError(
-                "batch_cache_read_per_mtok must be a finite non-negative "
-                f"Decimal or None, not {rate!r}"
-            )
+        for name in ("batch_cache_read_per_mtok", "batch_cache_write_5m_per_mtok"):
+            rate = getattr(self, name)
+            if rate is not None and not (
+                isinstance(rate, Decimal) and rate.is_finite() and rate >= 0
+            ):
+                raise ValueError(
+                    f"{name} must be a finite non-negative Decimal or None, not {rate!r}"
+                )
 
 
 @dataclass(frozen=True, slots=True)
@@ -370,6 +372,7 @@ def _billed_terms(resolved: ResolvedPrice) -> tuple[Any, ...]:
         price.batch_input_per_mtok,
         price.batch_output_per_mtok,
         price.batch_cache_read_per_mtok,
+        price.batch_cache_write_5m_per_mtok,
         price.request_fee_per_1k_usd,
         resolved.rules,
     )
@@ -523,6 +526,7 @@ def _price_tokens(
     input_rate = price.input_per_mtok
     output_rate = price.output_per_mtok
     cache_read_rate = price.cache_read_per_mtok
+    cache_write_5m_rate = price.cache_write_5m_per_mtok
     if batch:
         if price.batch_input_per_mtok is None or price.batch_output_per_mtok is None:
             reasons.append("batch_rate_unavailable")
@@ -533,6 +537,8 @@ def _price_tokens(
         # the standard rate, as it always has.
         if price.batch_cache_read_per_mtok is not None:
             cache_read_rate = price.batch_cache_read_per_mtok
+        if price.batch_cache_write_5m_per_mtok is not None:
+            cache_write_5m_rate = price.batch_cache_write_5m_per_mtok
 
     input_multiplier = Decimal("1")
     output_multiplier = Decimal("1")
@@ -579,12 +585,12 @@ def _price_tokens(
                 / _MILLION
             )
     if cache_write_5m_count:
-        if price.cache_write_5m_per_mtok is None:
+        if cache_write_5m_rate is None:
             reasons.append("cache_write_5m_rate_unavailable")
         else:
             cache_write_usd += (
                 Decimal(cache_write_5m_count)
-                * price.cache_write_5m_per_mtok
+                * cache_write_5m_rate
                 * input_multiplier
                 / _MILLION
             )

{metergraph_core-0.7.4 → metergraph_core-0.7.6}/src/metergraph_core/catalog_delta.py

@@ -46,6 +46,7 @@ RATE_FIELDS = (
     "batch_input_per_mtok",
     "batch_output_per_mtok",
     "batch_cache_read_per_mtok",
+    "batch_cache_write_5m_per_mtok",
     "request_fee_per_1k_usd",
 )
 _PRICE_FIELDS = frozenset(

{metergraph_core-0.7.4 → metergraph_core-0.7.6}/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: "2026-10-25"
+version: "2026-10-27"
 currency: USD
-pricing_verified_at: "2026-10-25"
+pricing_verified_at: "2026-10-27"
 models:
   - canonical_id: openai/gpt-6-sol
     publisher: openai
@@ -290,6 +290,30 @@ models:
           long_context: {threshold: 272000, input_multiplier: 2, output_multiplier: 1.5}
         source_url: https://ai-gateway.vercel.sh/v1/models
 
+  - canonical_id: openai/gpt-6.1-terra
+    publisher: openai
+    aliases:
+      - {provider: openai, alias: gpt-6.1-terra, channel: openai-api}
+      - {provider: litellm, alias: gpt-6.1-terra, channel: openai-api}
+      - {provider: unknown, alias: gpt-6.1-terra, channel: openai-api}
+    prices:
+      - channel: openai-api
+        region: global
+        effective_from: "2026-10-26"
+        input_per_mtok: 1.00
+        output_per_mtok: 5.00
+        cache_read_per_mtok: 0.05
+        cache_write_5m_per_mtok: 1.25
+        batch_input_per_mtok: 0.50
+        batch_output_per_mtok: 2.50
+        batch_cache_read_per_mtok: 0.025
+        batch_cache_write_5m_per_mtok: 0.625
+        rules:
+          input_includes_cache_read: true
+          input_includes_cache_write: true
+          long_context: {threshold: 272000, input_multiplier: 2, output_multiplier: 1.5}
+        source_url: https://developers.openai.com/api/docs/models/gpt-6.1-terra
+
   - canonical_id: openai/gpt-5.6-sol
     publisher: openai
     aliases:
@@ -3164,6 +3188,40 @@ models:
         rules:
           varies_by_provider: true
         source_url: https://openrouter.ai/api/v1/models
+  - canonical_id: moonshotai/kimi-k3.1-pro
+    publisher: moonshotai
+    aliases:
+      - {provider: moonshotai, alias: kimi-k3.1-pro, channel: moonshot-api}
+      - {provider: unknown, alias: kimi-k3.1-pro, channel: moonshot-api}
+      - {provider: litellm, alias: kimi-k3.1-pro, channel: moonshot-api}
+    prices:
+      - channel: moonshot-api
+        region: global
+        effective_from: "2026-10-24"
+        input_per_mtok: 2.00
+        output_per_mtok: 10.00
+        cache_read_per_mtok: 0.20
+        cache_write_5m_per_mtok: 2.00
+        rules: {input_includes_cache_read: true, input_includes_cache_write: true}
+        source_url: https://platform.kimi.ai/
+
+  - canonical_id: moonshotai/kimi-k3.1-pro-highspeed
+    publisher: moonshotai
+    aliases:
+      - {provider: moonshotai, alias: kimi-k3.1-pro-highspeed, channel: moonshot-api}
+      - {provider: unknown, alias: kimi-k3.1-pro-highspeed, channel: moonshot-api}
+      - {provider: litellm, alias: kimi-k3.1-pro-highspeed, channel: moonshot-api}
+    prices:
+      - channel: moonshot-api
+        region: global
+        effective_from: "2026-10-24"
+        input_per_mtok: 4.00
+        output_per_mtok: 20.00
+        cache_read_per_mtok: 0.20
+        cache_write_5m_per_mtok: 4.00
+        rules: {input_includes_cache_read: true, input_includes_cache_write: true}
+        source_url: https://platform.kimi.ai/
+
   - canonical_id: moonshotai/kimi-k3
     publisher: moonshotai
     aliases:

{metergraph_core-0.7.4 → metergraph_core-0.7.6}/src/metergraph_core/data/sources.yaml

@@ -152,6 +152,16 @@ sources:
     fingerprint: null
     low_risk: {eligible: false, dimensions: []}
 
+  - id: moonshot-api-pricing
+    provider: moonshotai
+    channel: moonshot-api
+    catalog_channel: true
+    url: https://platform.kimi.ai/
+    parser: html-price-prose
+    dimensions: [input_per_mtok, output_per_mtok, cache_read_per_mtok, cache_write_5m_per_mtok]
+    fingerprint: null
+    low_risk: {eligible: false, dimensions: []}
+
   # Gemini and partner models on Vertex AI. Prices vary by model, context
   # length, tier and processing mode; reviewed discovery only. The URL is the
   # current destination of Google's Vertex AI generative pricing page

{metergraph_core-0.7.4 → metergraph_core-0.7.6}/src/metergraph_core/extractors.py

@@ -7,8 +7,8 @@ the parser id a source registry entry names.
 Official parsers (OpenAI, Anthropic, Google, DeepSeek, Mistral, xAI and
 Perplexity) turn a page into ``{model name: {field: Decimal}}`` tables keyed
 by the page's own model spelling; :func:`extract` maps those to catalog
-price columns. The two generic discovery parsers, ``html-tables`` and
-``gpu-license-prose``, record normalized evidence from any page for a
+price columns. The generic discovery parsers, ``html-tables``,
+``html-price-prose``, and ``gpu-license-prose`` record normalized evidence for a
 reviewer and never produce applicable price rows; the license parser's
 evidence is GPU or license pricing, never token rates.
 
@@ -886,6 +886,23 @@ def _table_features(content: str) -> Any:
     return {"headers": sorted(list(row) for row in headers)}
 
 
+def _price_prose_evidence(content: str) -> dict[str, Any]:
+    text = page_text(content)
+    mentions = []
+    amounts = []
+    for match in _AMOUNT.finditer(text):
+        amounts.append(match.group()[1:].replace(",", ""))
+        excerpt = text[max(0, match.start() - 300):match.end() + 100].strip(" |")
+        if excerpt not in mentions:
+            mentions.append(excerpt)
+    return {"kind": "price_mentions", "mentions": mentions[:100], "amounts": amounts[:100]}
+
+
+def _price_prose_features(content: str) -> Any:
+    text = _AMOUNT.sub("$#", page_text(content))
+    return {"has_prices": "$#" in text, "has_model": "model" in text.lower()}
+
+
 _LICENSE_CANDIDATES = {
     "license_per_gpu_year_usd": re.compile(r"\$([0-9][0-9,]*(?:\.[0-9]+)?) per GPU per year"),
     "gpu_hour_usd": re.compile(r"\$([0-9][0-9,]*(?:\.[0-9]+)?) per GPU per hour"),
@@ -964,6 +981,12 @@ def redact_evidence(evidence: Mapping[str, Any]) -> dict[str, Any]:
             "statement_count": len(evidence.get("statements") or []),
             "candidates": dict(evidence.get("candidates") or {}),
         }
+    if kind == "price_mentions":
+        return {
+            "kind": kind,
+            "redacted": True,
+            "amounts": list(evidence.get("amounts") or []),
+        }
     return {"kind": kind, "redacted": True}
 
 
@@ -1006,6 +1029,7 @@ PARSERS: Mapping[str, ParserSpec] = MappingProxyType(
         # Generic discovery parsers: any page's tables, or GPU and license
         # pricing stated in prose. They describe no provider's layout.
         "html-tables": ParserSpec(False, _table_features, evidence=_table_evidence),
+        "html-price-prose": ParserSpec(False, _price_prose_features, evidence=_price_prose_evidence),
         "gpu-license-prose": ParserSpec(False, _license_features, evidence=_license_evidence),
     }
 )

{metergraph_core-0.7.4 → metergraph_core-0.7.6}/src/metergraph_core/loader.py

@@ -399,6 +399,15 @@ def parse_catalog(
                     f"{canonical}: invalid batch_cache_read_per_mtok "
                     f"{batch_cache_read!r}"
                 )
+            batch_cache_write = price.get("batch_cache_write_5m_per_mtok")
+            batch_cache_write_5m_per_mtok = _decimal(batch_cache_write)
+            if batch_cache_write is not None and (
+                batch_cache_write_5m_per_mtok is None or batch_cache_write_5m_per_mtok < 0
+            ):
+                raise CatalogError(
+                    f"{canonical}: invalid batch_cache_write_5m_per_mtok "
+                    f"{batch_cache_write!r}"
+                )
             prices.append(
                 Price(
                     id=f"{canonical}:{channel}:{region}:{_id_date(effective_from)}",
@@ -424,6 +433,7 @@ def parse_catalog(
                     publisher=publisher,
                     currency=currency,
                     batch_cache_read_per_mtok=batch_cache_read_per_mtok,
+                    batch_cache_write_5m_per_mtok=batch_cache_write_5m_per_mtok,
                 )
             )
     return version, {key: tuple(value) for key, value in aliases.items()}, _unique_ids(prices)

{metergraph_core-0.7.4 → metergraph_core-0.7.6/src/metergraph_core.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: metergraph-core
-Version: 0.7.4
+Version: 0.7.6
 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
@@ -186,6 +186,8 @@ standard `cache_read_per_mtok`, as before. A standard call always uses the
 standard cache-read rate. A row with only a batch cache rate fully prices a
 cached batch call, while a cached standard call on it is `partial` with
 `cache_read_rate_unavailable`.
+Batch cache writes use `Price.batch_cache_write_5m_per_mtok` when stated;
+otherwise they retain the row's standard cache-write rate.
 
 A `ModelUsage` count left `None` is unknown, not zero. A missing input or
 output count keeps the subtotal that is known, marks the result `partial`, and