@linxiraos/pi-catalog 1.1.26 → 1.2.0

package/src/compat/rules/providers/openai-codex.kdl

@@ -3,6 +3,7 @@
 provider "openai-codex" {
 	default-model "gpt-5.5"
 	env "OPENAI_CODEX_OAUTH_TOKEN"
+	dynamic-models-authoritative #true
 	kind-apis {
 		image "openai-codex-responses"
 	}
@@ -24,9 +25,13 @@ provider "openai-codex" {
 	}
 
 	// Replaces the Codex provider's handwritten flex/priority pricing fallback.
+	// Fast (priority) draws included subscription usage at 2.5x Standard on
+	// every model that supports it; purchased credits and Enterprise
+	// pay-as-you-go bill 2x. See
+	// https://learn.chatgpt.com/docs/agent-configuration/speed.
 	service-tier-cost {
 		flex 0.5
-		priority 2
+		priority 2.5
 	}
 	// Harmony-protocol leak detection/mitigation applies to every Codex model,
 	// current and future; replaces the provider check in harmony-leak.ts.
@@ -72,10 +77,6 @@ provider "openai-codex" {
 			cache-read 1.0
 			cache-write 0
 		}
-		service-tier-cost {
-			flex 0.5
-			priority 2.5
-		}
 		limits-patch {
 			context-window 272000
 		}
@@ -106,6 +107,18 @@ provider "openai-codex" {
 		}
 		max-context-window 922000
 	}
+	// Codex Ultrafast ships on GPT-6 Astra and GPT-6.1 Sol. This table uses
+	// OpenAI's included-usage multipliers for Codex: Ultrafast 8x and Fast 2.5x
+	// Standard (purchased credits / Enterprise pay-as-you-go bill 6x and 2x
+	// instead). See https://learn.chatgpt.com/docs/agent-configuration/speed
+	// and https://learn.chatgpt.com/docs/pricing.
+	models "gpt-6-astra" "gpt-6-astra-wm" "gpt-6.1-sol" "gpt-6.1-sol-wm" {
+		service-tier-cost {
+			flex 0.5
+			priority 2.5
+			ultrafast 8
+		}
+	}
 	models "gpt-6-luna" "gpt-6-luna-wm" {
 		cost-patch {
 			input 0.1
@@ -126,14 +139,6 @@ provider "openai-codex" {
 		revision "<5.4" {
 			supports-reasoning-summary #false
 		}
-		// Subscription Codex rates the priority service tier at 2.5x for the
-		// GPT-5.5 generation (2x elsewhere).
-		revision "=5.5" {
-			service-tier-cost {
-				flex 0.5
-				priority 2.5
-			}
-		}
 		// Codex discovery reports inconsistent priorities for the GPT-5.4
 		// family; normalize by variant. Mini/nano also keep the lower prompt
 		// budget despite their plain OpenAI ids.

package/src/compat/rules/providers/openai.kdl

@@ -84,8 +84,23 @@ provider "openai" {
 			limits context=8192
 			supports-tools #false
 		}
+		// Decisions API (`POST /v1/decisions`, public beta,
+		// developers.openai.com/api/docs/guides/decisions) answers typed questions
+		// with GPT-6 Luna. The wire id is the chat row's `gpt-6-luna`, so the judge
+		// row takes its own id. Billed $0.10 per 1M input tokens, no output charge.
+		model "gpt-6-luna-decisions" name="GPT-6 Luna Decisions" api="openai-decisions" base-url="https://api.openai.com/v1" {
+			request-model-id "gpt-6-luna"
+			reasoning #false
+			input "text" "image"
+			cost input=0.1 output=0 cache-read=0 cache-write=0
+			limits context=1050000
+			supports-tools #false
+		}
 	}
 
+	models "gpt-6-luna-decisions" {
+		kind "judge"
+	}
 	models "whisper-1" "gpt-4o-transcribe" "gpt-4o-mini-transcribe" {
 		kind "stt"
 	}
@@ -177,6 +192,13 @@ provider "openai" {
 			cache-write 25.0
 		}
 	}
+	// API Ultrafast is 6x the standard card on both context tiers: GPT-6 Astra
+	// $60/$6/$75/$300 and GPT-6.1 Sol $12/$0.60/$15/$60 per 1M short-context.
+	models "gpt-6-astra*" "gpt-6.1-sol*" {
+		service-tier-cost {
+			ultrafast 6
+		}
+	}
 	// residue: the bare Daybreak rolling aliases pin the documented 5.6-gen
 	// wire restrictions without carrying the gpt- prefix identity.
 	models "daybreak-blue-latest" "daybreak-red-latest" {

package/src/compat/rules/providers/opencode-go.kdl

@@ -78,6 +78,17 @@ provider "opencode-go" {
 			thinking-mode "effort"
 		}
 	}
+	// LongCat exposes a reasoning toggle, not selectable effort tiers.
+	// Off uses the gateway's boolean reasoning switch, never a fake lowest tier.
+	models "longcat-2.5-preview-free" {
+		thinking-efforts
+		supports-reasoning-effort #false
+		reasoning-disable-mode "openrouter-enabled-false"
+	}
+	// Space Bunny publishes low through max, without the generic minimal tier.
+	models "space-bunny-free" {
+		thinking-efforts "low" "medium" "high" "xhigh" "max"
+	}
 	// residue: taxonomy ranks and exact globs do not isolate these models.
 	models "glm-5" "glm-5.1" {
 		thinking-efforts "minimal" "low" "medium" "high" "xhigh"

package/src/compat/rules/providers/opencode-zen.kdl

@@ -116,4 +116,24 @@ provider "opencode-zen" {
 		thinking-efforts "minimal" "low" "medium" "high"
 		thinking-mode "effort"
 	}
+	// Zen serves Jev over System One, not chat completions.
+	seed api="typesafe" bundle="always" base-url="https://opencode.ai/zen" {
+		model "jev-1.13" name="Jev 1.13" {
+			reasoning #false
+			input "text"
+			cost input=0.042 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+		model "jev-1.13-free" name="Jev 1.13 Free" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+	}
+	models "jev-1.13" "jev-1.13-free" {
+		kind "judge"
+	}
 }

package/src/compat/rules/providers/openrouter.kdl

@@ -159,6 +159,12 @@ provider "openrouter" {
 		family "o-series" {
 			thinking-efforts "minimal" "low" "medium" "high"
 		}
+		// OpenRouter's Responses API does not document `configuration_update`
+		// input items (and serves `-pro` rows that would need pro mode, where the
+		// item is unsupported): GPT-6 effort changes stay on the request level.
+		revision ">=6 <7" {
+			supports-configuration-update #false
+		}
 	}
 	class "stepfun" {
 		family "step" {

package/src/compat/rules/providers/snowflake.kdl

@@ -0,0 +1,135 @@
+// Documented account-billed Cortex REST API, not verified CoCo-subscription usage.
+// Model IDs and compatibility: https://docs.snowflake.com/en/user-guide/snowflake-cortex/cortex-rest-api
+// Output is capped at the Cortex REST API's documented 16,384-token maximum (rate-limit table;
+// newer models absent from that table are assumed to share it). Context windows are conservative
+// AI_COMPLETE configured ceilings, not verified REST maxima:
+// https://docs.snowflake.com/en/user-guide/snowflake-cortex/cortex-rest-api#label-cortex-rest-api-rate-limits
+// https://docs.snowflake.com/en/user-guide/snowflake-cortex/aisql-regional-availability
+// USD per million tokens estimates: Consumption Table 6(b) AI Credits x $2.00
+// (on-demand global routing); regional routing and contract discounts change billing.
+// https://www.snowflake.com/legal-files/CreditConsumptionTable.pdf
+// https://docs.snowflake.com/en/user-guide/snowflake-cortex/pricing
+provider "snowflake" {
+	default-model "claude-opus-5-5"
+	env "SNOWFLAKE_PAT"
+
+	// Credential-scoped roster: never fetch during catalog generation. The `.invalid`
+	// origin is a placeholder that snowflakeTransport replaces with the account URL.
+	seed api="anthropic-messages" base-url="https://snowflake-account.invalid/api/v2/cortex" bundle="always" precedence="seed" {
+		model "claude-opus-5-5" name="Claude Opus 5.5" {
+			reasoning #true
+			input "text" "image"
+			cost input=4 output=20 cache-write=5 cache-read=0.2
+			limits context=1000000 max-tokens=16384
+		}
+		model "claude-opus-5" name="Claude Opus 5" {
+			reasoning #true
+			input "text" "image"
+			cost input=5 output=25 cache-write=6.25 cache-read=0.5
+			limits context=1000000 max-tokens=16384
+		}
+		model "claude-opus-4-8" name="Claude Opus 4.8" {
+			reasoning #true
+			input "text" "image"
+			cost input=5 output=25 cache-write=6.25 cache-read=0.5
+			limits context=1000000 max-tokens=16384
+		}
+		model "claude-opus-4-7" name="Claude Opus 4.7" {
+			reasoning #true
+			input "text" "image"
+			cost input=5 output=25 cache-write=6.25 cache-read=0.5
+			limits context=1000000 max-tokens=16384
+		}
+		model "claude-opus-4-6" name="Claude Opus 4.6" {
+			reasoning #true
+			input "text" "image"
+			cost input=5 output=25 cache-write=6.25 cache-read=0.5
+			limits context=1000000 max-tokens=16384
+		}
+		model "claude-opus-4-5" name="Claude Opus 4.5" {
+			reasoning #true
+			input "text" "image"
+			cost input=5 output=25 cache-write=6.25 cache-read=0.5
+			limits context=200000 max-tokens=16384
+		}
+		model "claude-sonnet-5" name="Claude Sonnet 5" {
+			reasoning #true
+			input "text" "image"
+			cost input=2 output=10 cache-write=2.5 cache-read=0.2
+			limits context=1000000 max-tokens=16384
+		}
+		model "claude-sonnet-4-6" name="Claude Sonnet 4.6" {
+			reasoning #true
+			input "text" "image"
+			cost input=3 output=15 cache-write=3.75 cache-read=0.3
+			limits context=1000000 max-tokens=16384
+		}
+		model "claude-sonnet-4-5" name="Claude Sonnet 4.5" {
+			reasoning #true
+			input "text" "image"
+			cost input=3 output=15 cache-write=3.75 cache-read=0.3
+			limits context=200000 max-tokens=16384
+		}
+		model "claude-haiku-4-5" name="Claude Haiku 4.5" {
+			reasoning #true
+			input "text" "image"
+			cost input=1 output=5 cache-write=1.25 cache-read=0.1
+			limits context=200000 max-tokens=16384
+		}
+		model "openai-gpt-5.1" name="GPT-5.1" api="openai-completions" base-url="https://snowflake-account.invalid/api/v2/cortex/v1" {
+			reasoning #true
+			input "text" "image"
+			cost input=1.25 output=10 cache-write=0 cache-read=0.13
+			limits context=272000 max-tokens=16384
+		}
+		model "openai-gpt-5" name="GPT-5" api="openai-completions" base-url="https://snowflake-account.invalid/api/v2/cortex/v1" {
+			reasoning #true
+			input "text" "image"
+			cost input=1.25 output=10 cache-write=0 cache-read=0.13
+			limits context=272000 max-tokens=16384
+		}
+		model "openai-gpt-5-mini" name="GPT-5 Mini" api="openai-completions" base-url="https://snowflake-account.invalid/api/v2/cortex/v1" {
+			reasoning #true
+			input "text" "image"
+			cost input=0.25 output=2 cache-write=0 cache-read=0.03
+			limits context=272000 max-tokens=16384
+		}
+		model "openai-gpt-5-nano" name="GPT-5 Nano" api="openai-completions" base-url="https://snowflake-account.invalid/api/v2/cortex/v1" {
+			reasoning #true
+			input "text" "image"
+			cost input=0.06 output=0.44 cache-write=0 cache-read=0.01
+			limits context=272000 max-tokens=16384
+		}
+		model "openai-gpt-4.1" name="GPT-4.1" api="openai-completions" base-url="https://snowflake-account.invalid/api/v2/cortex/v1" {
+			reasoning #false
+			input "text" "image"
+			cost input=2 output=8 cache-write=0 cache-read=0.5
+			limits context=128000 max-tokens=16384
+		}
+	}
+
+	// Claude thinking modes and effort ladders come from the Anthropic lineage rules
+	// (classes/anthropic.kdl). Live Cortex probes matched them: Opus 4.7+/Sonnet 5
+	// accept xhigh and max; Opus 4.6 and Sonnet 4.6 reject xhigh.
+	// Cortex compatibility chart: ephemeral 5-minute caching and Bedrock-compatible betas only.
+	class "anthropic" {
+		supports-eager-tool-input-streaming #false
+		supports-long-cache-retention #false
+	}
+	// Cortex documents reasoning_effort none/minimal/low/medium/high. Prefixed ids
+	// (openai-gpt-*) carry no `gpt` family, so the <5.2 lineage ladder is restated
+	// here; 5.2+ rows fall through to the unscoped lineage ladders in classes/openai.kdl.
+	class "openai" {
+		revision ">=5 <5.2" {
+			thinking-mode "effort"
+			thinking-efforts "minimal" "low" "medium" "high"
+		}
+	}
+	// Cortex compatibility chart: max_tokens/store are hard errors; non-OpenAI backends accept user/assistant/system roles only.
+	on-api "openai-completions" {
+		max-tokens-field "max_completion_tokens"
+		supports-store #false
+		supports-developer-role #false
+		supports-reasoning-effort #true
+	}
+}

package/src/compat/rules/providers/tool-result-images.kdl

@@ -0,0 +1,16 @@
+// Converse rejects image blocks nested inside toolResult content for OpenAI models,
+// so their tool-result images move into sibling user blocks. The rule is scoped by
+// adapter, not provider name, so a custom provider on bedrock-converse-stream gets
+// it too.
+on-api "bedrock-converse-stream" {
+	class "openai" {
+		requires-tool-result-image-hoisting #true
+	}
+	// Opaque application-inference-profile ARNs carry no vendor signal, so an
+	// OpenAI-backed profile classifies here and would otherwise nest images in
+	// toolResult content and hit the same rejection. Hoisting is accepted by
+	// every Converse model measured, so it is the safe default for unknowns.
+	class "unknown" {
+		requires-tool-result-image-hoisting #true
+	}
+}

package/src/compat/rules/providers/zhipu-coding-plan.kdl

@@ -5,6 +5,16 @@ provider "zhipu-coding-plan" {
 	env "ZHIPU_API_KEY"
 	dynamic-models-authoritative #true
 	discovery label="Zhipu Coding Plan"
+	// The coding-plan upstream snapshot omits FlashX although the served SKU
+	// has the same limits and mandatory-thinking ladder as GLM-5.3-Flash.
+	seed api="openai-completions" base-url="https://open.bigmodel.cn/api/coding/paas/v4" bundle="always" {
+		model "glm-5.3-flashx" name="GLM-5.3-FlashX" {
+			reasoning #true
+			input "text" "image"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits context=1000000 max-tokens=131072
+		}
+	}
 
 	// Replaces the Zhipu provider thinking-format branch.
 	thinking-format "zai"

package/src/compat/rules/runtime/behavior.kdl

@@ -92,13 +92,13 @@ behavior {
 	// Factory Droid bills each model from one subscription pool: Droid Core
 	// (flat rate) or Standard Credits. Unknown ids have no inferred entitlement.
 	quota-tiers provider="factory-droid" {
-		tier "core" "inkling" "mistral-medium-3.5" "glm-5.3-flash" "glm-5.3" "glm-5.2" "glm-5.2-fast" \
+		tier "core" "inkling" "mistral-large-4" "mistral-medium-3.5" "glm-5.3-flash" "glm-5.3" "glm-5.2" "glm-5.2-fast" \
 			"kimi-k3" "qwen3.8-max" "nemotron-3-ultra" "deepseek-v4.1-flash" "deepseek-v4-flash-0731" \
 			"deepseek-v4-pro" "minimax-m3" "minimax-m2.7"
 		tier "standard" "claude-fable-5.1" "claude-fable-5" "claude-opus-5-5" "claude-opus-5-5-fast" \
 			"claude-opus-5" "claude-opus-5-fast" "claude-opus-4-8" "claude-opus-4-8-fast" "claude-opus-4-7" \
 			"claude-opus-4-6" "claude-opus-4-5-20251101" "claude-sonnet-5-5" "claude-sonnet-5" \
-			"claude-sonnet-4-6" "claude-sonnet-4-5-20250929" "claude-haiku-4-5-20251001" "gpt-6-astra" \
+			"claude-sonnet-4-6" "claude-sonnet-4-5-20250929" "claude-haiku-5-5" "claude-haiku-4-5-20251001" "gpt-6-astra" \
 			"gpt-6.1-sol" "gpt-6-sol" "gpt-6-luna" "gpt-5.6-sol" "gpt-5.6-sol-fast" "gpt-5.6-terra" \
 			"gpt-5.6-luna" "gpt-5.5" "gpt-5.5-fast" "gpt-5.5-pro" "gpt-5.4" "gpt-5.4-fast" "gpt-5.4-mini" \
 			"gpt-5.4-mini-fast" "gpt-5.3-codex" "gpt-5.3-codex-fast" "gpt-5.2" "garnet-07-15" \
@@ -172,7 +172,7 @@ behavior {
 	// SuperGrok's roster interleaves media models the chat picker cannot
 	// serve; the dedicated tool surfaces route them instead.
 	exclude-models provider="xai-oauth" exact="grok-imagine-image-quality" exact="grok-imagine-image-2.0" \
-		exact="grok-imagine-video" exact="grok-imagine-video-1.5" prefix="grok-stt-" prefix="grok-voice-"
+		prefix="grok-imagine-video" prefix="grok-stt-" prefix="grok-voice-"
 	// Meta's /v1/models lists image generation and transcription SKUs beside
 	// the Muse Spark chat models.
 	exclude-models provider="meta" prefix="muse-image-" prefix="muse-voice-"
@@ -279,7 +279,8 @@ behavior {
 			exact="claude-opus-4-8" exact="claude-opus-4-8-fast" exact="claude-opus-4-7" \
 			exact="claude-opus-4-6" exact="claude-opus-4-5-20251101" exact="claude-sonnet-5-5" \
 			exact="claude-sonnet-5" exact="claude-sonnet-4-6" exact="claude-sonnet-4-5-20250929" \
-			exact="claude-haiku-4-5-20251001" exact="atlas-07-21" exact="aster-07-15" exact="minimax-m2.7"
+			exact="claude-haiku-5-5" exact="claude-haiku-4-5-20251001" exact="atlas-07-21" exact="aster-07-15" \
+			exact="minimax-m2.7"
 		route "openai-responses" exact="gpt-6-astra" exact="gpt-6.1-sol" exact="gpt-6-sol" exact="gpt-6-luna" \
 			exact="gpt-5.6-sol" exact="gpt-5.6-sol-fast" exact="gpt-5.6-terra" exact="gpt-5.6-luna" \
 			exact="gpt-5.5" exact="gpt-5.5-fast" exact="gpt-5.5-pro" exact="gpt-5.4" exact="gpt-5.4-fast" \
@@ -288,7 +289,7 @@ behavior {
 		route "google-generate" exact="garnet-07-15" exact="gemini-3.1-pro-preview" exact="gemini-3.8-flash" \
 			exact="gemini-3.7-flash" exact="gemini-3.6-flash" exact="gemini-3.5-flash" \
 			exact="gemini-3-flash-preview"
-		route "openai-completions" exact="inkling" exact="mistral-medium-3.5" exact="glm-5.3-flash" \
+		route "openai-completions" exact="inkling" exact="mistral-large-4" exact="mistral-medium-3.5" exact="glm-5.3-flash" \
 			exact="glm-5.3" exact="glm-5.2" exact="glm-5.2-fast" exact="kimi-k3" exact="qwen3.8-max" \
 			exact="nemotron-3-ultra" exact="deepseek-v4.1-flash" exact="deepseek-v4-flash-0731" \
 			exact="deepseek-v4-pro" exact="minimax-m3"
@@ -351,6 +352,8 @@ behavior {
 		alias "claude-opus-4-6" peer-id="claude-opus-4-6@default"
 		alias "claude-sonnet-4-5" peer-id="claude-sonnet-4-5@20250929"
 		alias "claude-sonnet-4-6" peer-id="claude-sonnet-4-6@default"
+		alias "claude-opus-5-5" peer-id="claude-opus-5-5@default"
+		alias "claude-sonnet-5-5" peer-id="claude-sonnet-5-5@default"
 	}
 	// SuperGrok surfaces a few models under IDs that differ from their public
 	// xai equivalent, so the exact-id price mirror misses them.

package/src/compat/rules/taxonomy/_collapse.kdl

@@ -88,6 +88,24 @@ collapse {
 		mode "budget"
 		efforts "minimal" "low" "medium" "high"
 	}
+	variant-family "google-antigravity" "claude-opus-5-5" name="Claude Opus 5.5" {
+		members "claude-opus-5-5-low" "claude-opus-5-5-medium" "claude-opus-5-5-high"
+		route "low" "claude-opus-5-5-low"
+		route "medium" "claude-opus-5-5-medium"
+		route "high" "claude-opus-5-5-high"
+		mode "budget"
+		efforts "low" "medium" "high"
+		requires-effort #true
+	}
+	variant-family "google-antigravity" "claude-sonnet-5-5" name="Claude Sonnet 5.5" {
+		members "claude-sonnet-5-5-low" "claude-sonnet-5-5-medium" "claude-sonnet-5-5-high"
+		route "low" "claude-sonnet-5-5-low"
+		route "medium" "claude-sonnet-5-5-medium"
+		route "high" "claude-sonnet-5-5-high"
+		mode "budget"
+		efforts "low" "medium" "high"
+		requires-effort #true
+	}
 	variant-family "google-antigravity" "claude-sonnet-4-6" name="Claude Sonnet 4.6" {
 		members "claude-sonnet-4-6" "claude-sonnet-4-6-thinking"
 		mode "budget"
@@ -639,6 +657,12 @@ collapse {
 	// Retired id: the Codex image model was relabeled to what the backend runs
 	// (gpt-image-2-codex); keep saved `openai-codex/gpt-image-1` roles resolving.
 	provider-alias "openai-codex" "gpt-image-1" "gpt-image-2"
+	// Retired rich-lane ids: AvailableModels discovery named Grok 4.5/4.6 lanes
+	// after the unprefixed model name before they adopted these family ids
+	// (#14164). Family `aliases` (not `provider-alias`) so saved selectors and
+	// models.yml `modelOverrides` keyed by the old ids both migrate. Grok 4.6
+	// is spelled out instead of riding the `cursor-grok-{rev}` template: a
+	// templated `grok-{rev}` alias would also claim the live Grok 4.7+ ids.
 	variant-family "cursor" "cursor-grok-4.5" name="Grok 4.5" {
 		members "cursor-grok-4.5-low" "cursor-grok-4.5-medium" "cursor-grok-4.5-high"
 		route "low" "cursor-grok-4.5-low"
@@ -648,6 +672,7 @@ collapse {
 		efforts "low" "medium" "high"
 		requires-effort #true
 		default-member "cursor-grok-4.5-medium"
+		aliases "grok-4.5"
 	}
 	variant-family "cursor" "cursor-grok-4.5-fast" name="Grok 4.5 Fast" {
 		members "cursor-grok-4.5-low-fast" "cursor-grok-4.5-medium-fast" "cursor-grok-4.5-high-fast"
@@ -658,8 +683,33 @@ collapse {
 		efforts "low" "medium" "high"
 		requires-effort #true
 		default-member "cursor-grok-4.5-medium-fast"
+		aliases "grok-4.5-fast"
+	}
+	variant-family "cursor" "cursor-grok-4.6" name="Grok 4.6" {
+		members "cursor-grok-4.6-low" "cursor-grok-4.6-medium" "cursor-grok-4.6-high" "cursor-grok-4.6-xhigh"
+		route "low" "cursor-grok-4.6-low"
+		route "medium" "cursor-grok-4.6-medium"
+		route "high" "cursor-grok-4.6-high"
+		route "xhigh" "cursor-grok-4.6-xhigh"
+		mode "effort"
+		efforts "low" "medium" "high" "xhigh"
+		requires-effort #true
+		default-member "cursor-grok-4.6-medium"
+		aliases "grok-4.6"
+	}
+	variant-family "cursor" "cursor-grok-4.6-fast" name="Grok 4.6 Fast" {
+		members "cursor-grok-4.6-low-fast" "cursor-grok-4.6-medium-fast" "cursor-grok-4.6-high-fast" "cursor-grok-4.6-xhigh-fast"
+		route "low" "cursor-grok-4.6-low-fast"
+		route "medium" "cursor-grok-4.6-medium-fast"
+		route "high" "cursor-grok-4.6-high-fast"
+		route "xhigh" "cursor-grok-4.6-xhigh-fast"
+		mode "effort"
+		efforts "low" "medium" "high" "xhigh"
+		requires-effort #true
+		default-member "cursor-grok-4.6-medium-fast"
+		aliases "grok-4.6-fast"
 	}
-	variant-family "cursor" "cursor-grok-{rev}" name="Grok {rev}" revision=">=4.6" {
+	variant-family "cursor" "cursor-grok-{rev}" name="Grok {rev}" revision=">=4.7" {
 		members "cursor-grok-{rev}-low" "cursor-grok-{rev}-medium" "cursor-grok-{rev}-high" "cursor-grok-{rev}-xhigh"
 		route "low" "cursor-grok-{rev}-low"
 		route "medium" "cursor-grok-{rev}-medium"
@@ -670,7 +720,7 @@ collapse {
 		requires-effort #true
 		default-member "cursor-grok-{rev}-medium"
 	}
-	variant-family "cursor" "cursor-grok-{rev}-fast" name="Grok {rev} Fast" revision=">=4.6" {
+	variant-family "cursor" "cursor-grok-{rev}-fast" name="Grok {rev} Fast" revision=">=4.7" {
 		members "cursor-grok-{rev}-low-fast" "cursor-grok-{rev}-medium-fast" "cursor-grok-{rev}-high-fast" "cursor-grok-{rev}-xhigh-fast"
 		route "low" "cursor-grok-{rev}-low-fast"
 		route "medium" "cursor-grok-{rev}-medium-fast"

package/src/compat/rules/taxonomy/deepseek.kdl

@@ -15,6 +15,10 @@ class "deepseek" {
     family "v4" glob="*deepseek-v4*"
     family "v3" glob="*deepseek-v3*"
 
+    revision prefix="deepseek-v" anywhere=#true
+
+    override id="deepseek-bare-flash-v4-1" provider="deepseek" model="deepseek-flash" class="deepseek" family="flash" revision="4.1" rationale="The first-party bare Flash alias serves DeepSeek V4.1 Flash, not V4 Flash" provenance="providers/deepseek.kdl"
+    override id="opencode-go-bare-flash-v4-1" provider="opencode-go" model="deepseek-flash" class="deepseek" family="flash" revision="4.1" rationale="OpenCode Go's bare Flash lane shares the reviewed V4.1 Flash deployment contract" provenance="providers/opencode-go.kdl"
     override id="vercel-deepseek-v3-2-thinking-product" provider="vercel-ai-gateway" model="deepseek-v3.2-thinking" logical="deepseek/deepseek-v3.2-thinking" class="deepseek" rationale="Oracle exposes this independently priced product rather than an effort route" provenance="fixtures/llm-oracle/catalog/models.normalized.json"
     override id="opencode-zen-big-pickle-family" provider="opencode-zen" model="big-pickle" class="deepseek" rationale="OpenCode Zen's big-pickle is a DeepSeek reasoning alias whose upstream 400s require exact reasoning_content replay" provenance="models.json census"
     override id="umans-deepseek-v4-flash-family" provider="umans" model="umans-deepseek-v4-flash-0731" logical="umans-deepseek-v4-flash-0731" class="deepseek" rationale="The opaque Umans product identifier is backed by the reviewed DeepSeek V4 Flash family" provenance="fixtures/llm-oracle/catalog-policy/compat-profiles.json:compat-04"

package/src/compat/rules/taxonomy/kimi.kdl

@@ -1,6 +1,9 @@
 class "kimi" {
     namespace "moonshotai"
     bounded "kimi"
+    // Kimi Code exposes bare K3 SKUs; selectors may include a provider namespace.
+    exact "k3"
+    exact "k3-256k"
 
     family "k2" glob="*kimi-k2*"
     family "k2.5" glob="*kimi-k2.5*" priority=10
@@ -13,6 +16,8 @@ class "kimi" {
     family "k2.7-code" glob="*kimi-k2p7*code*" priority=10
     family "k2-thinking" glob="*kimi-k2-thinking*" priority=20
     family "k3" glob="*kimi-k3*"
+    family "k3" glob="k3"
+    family "k3" glob="k3-256k"
 
     override id="kilo-kimi-k2-thinking-product" provider="kilo" model="kimi-k2-thinking" logical="moonshotai/kimi-k2-thinking" class="kimi" rationale="Oracle exposes this independently priced product rather than an effort route" provenance="fixtures/llm-oracle/catalog/models.normalized.json"
     override id="openrouter-kimi-k2-thinking-product" provider="openrouter" model="kimi-k2-thinking" logical="moonshotai/kimi-k2-thinking" class="kimi" rationale="Oracle exposes this independently priced product rather than an effort route" provenance="fixtures/llm-oracle/catalog/models.normalized.json"