@oh-my-pi/pi-catalog 18.6.1 → 18.7.0

package/src/compat/rules/classes/anthropic.kdl

@@ -1,5 +1,70 @@
 // Model-lineage compat for "anthropic"; regenerated from the frozen census using class taxonomy selectors.
 
+// Images bill ⌈w/28⌉·⌈h/28⌉ visual tokens on every host. Claude 4.7+ and
+// Fable/Mythos read up to 2576px and 4,784 tokens; older lines fit 1568px and
+// cap at 1,568 tokens, downscaling past it.
+// https://platform.claude.com/docs/en/build-with-claude/vision#evaluate-image-size
+// The class default is the high-res tier, so ids without a revision
+// (`claude-opus-latest`, `~anthropic/claude-sonnet-latest`) and future lines
+// are never undercounted; only Opus/Sonnet/Haiku revisions known to predate
+// 4.7 take the standard tier. Separator-collapsed revisions parse as two
+// digits (`claude-opus-45` is Opus 4.5 → 45), so 10–46 is standard too.
+class "anthropic" {
+	image-tokenization {
+		regime "anthropic-patch"
+		max-edge 2576
+		max-tokens 4784
+	}
+	family "opus" {
+		revision "<4.7" {
+			image-tokenization {
+				regime "anthropic-patch"
+				max-edge 1568
+				max-tokens 1568
+			}
+		}
+		revision ">=10 <47" {
+			image-tokenization {
+				regime "anthropic-patch"
+				max-edge 1568
+				max-tokens 1568
+			}
+		}
+	}
+	family "sonnet" {
+		revision "<4.7" {
+			image-tokenization {
+				regime "anthropic-patch"
+				max-edge 1568
+				max-tokens 1568
+			}
+		}
+		revision ">=10 <47" {
+			image-tokenization {
+				regime "anthropic-patch"
+				max-edge 1568
+				max-tokens 1568
+			}
+		}
+	}
+	family "haiku" {
+		revision "<4.7" {
+			image-tokenization {
+				regime "anthropic-patch"
+				max-edge 1568
+				max-tokens 1568
+			}
+		}
+		revision ">=10 <47" {
+			image-tokenization {
+				regime "anthropic-patch"
+				max-edge 1568
+				max-tokens 1568
+			}
+		}
+	}
+}
+
 class "anthropic" {
 	family "fable" {
 		revision ">=5.1" {
@@ -181,7 +246,7 @@ class "anthropic" {
 		thinking-efforts "minimal" "low" "medium" "high" "xhigh"
 	}
 	on "amazon-bedrock" "anthropic" "cloudflare-ai-gateway" "github-copilot" "gitlab-duo" \
-		"google-antigravity" "google-vertex" "opencode-zen" "vercel-ai-gateway" "zenmux" {
+		"google-antigravity" "google-vertex" "opencode-zen" "snowflake" "vercel-ai-gateway" "zenmux" {
 		family "sonnet" {
 			revision ">=3.7 <4.6" {
 				thinking-mode "budget"
@@ -189,13 +254,13 @@ class "anthropic" {
 		}
 	}
 	on "amazon-bedrock" "anthropic" "cloudflare-ai-gateway" "github-copilot" "gitlab-duo" \
-		"google-vertex" "opencode-zen" "vercel-ai-gateway" {
+		"google-vertex" "opencode-zen" "snowflake" "vercel-ai-gateway" {
 		family "haiku" {
 			thinking-mode "budget"
 		}
 	}
 	on "amazon-bedrock" "anthropic" "cloudflare-ai-gateway" "github-copilot" "gitlab-duo" \
-		"google-vertex" "opencode-zen" "vercel-ai-gateway" "zenmux" {
+		"google-vertex" "opencode-zen" "snowflake" "vercel-ai-gateway" "zenmux" {
 		family "opus" {
 			revision ">=4.5 <4.6" {
 				thinking-mode "anthropic-budget-effort"
@@ -203,7 +268,7 @@ class "anthropic" {
 		}
 	}
 	on "amazon-bedrock" "anthropic" "cloudflare-ai-gateway" "github-copilot" "google-vertex" \
-		"opencode-zen" "openrouter" "vercel-ai-gateway" "zenmux" {
+		"opencode-zen" "openrouter" "snowflake" "vercel-ai-gateway" "zenmux" {
 		family "opus" {
 			revision ">=4.6 <4.7" {
 				thinking-efforts "low" "medium" "high" "max"
@@ -211,7 +276,7 @@ class "anthropic" {
 		}
 	}
 	on "amazon-bedrock" "anthropic" "cloudflare-ai-gateway" "github-copilot" "google-vertex" \
-		"opencode-zen" "vercel-ai-gateway" "zenmux" {
+		"opencode-zen" "snowflake" "vercel-ai-gateway" "zenmux" {
 		revision ">=4.7 <45.1" {
 			thinking-mode "anthropic-adaptive"
 		}
@@ -242,13 +307,13 @@ class "anthropic" {
 		}
 	}
 	on "anthropic" "cloudflare-ai-gateway" "github-copilot" "google-vertex" "opencode-zen" \
-		"openrouter" "vercel-ai-gateway" "zenmux" {
+		"openrouter" "snowflake" "vercel-ai-gateway" "zenmux" {
 		revision ">=4.7 <45.1" {
 			thinking-efforts "low" "medium" "high" "xhigh" "max"
 		}
 	}
 	on "anthropic" "cloudflare-ai-gateway" "github-copilot" "google-vertex" "opencode-zen" \
-		"vercel-ai-gateway" "zenmux" {
+		"snowflake" "vercel-ai-gateway" "zenmux" {
 		family "sonnet" {
 			revision ">=4.6 <5" {
 				thinking-efforts "low" "medium" "high"

package/src/compat/rules/classes/gemini.kdl

@@ -118,3 +118,15 @@ class "gemini" {
 		}
 	}
 }
+
+// Gemini 3+ bills a fixed `media_resolution` budget per image regardless of
+// pixels; the default for images is HIGH, 1,120 tokens.
+// https://ai.google.dev/gemini-api/docs/media-resolution
+class "gemini" {
+	revision ">=3" {
+		image-tokenization {
+			regime "fixed"
+			tokens 1120
+		}
+	}
+}

package/src/compat/rules/classes/minimax.kdl

@@ -51,7 +51,7 @@ class "minimax" {
 			thinking-efforts "minimal" "low" "medium" "high" "xhigh"
 		}
 	}
-	on "minimax" "minimax-cn" "vercel-ai-gateway" {
+	on "minimax" "minimax-cn" "minimax-code" "minimax-code-cn" "vercel-ai-gateway" {
 		thinking-mode "anthropic-adaptive"
 		family "m2" {
 			// Replaces the Anthropic-route MiniMax effort remap table.
@@ -71,6 +71,26 @@ class "minimax" {
 			thinking-efforts "low" "medium" "high"
 		}
 	}
+	// MiniMax-M3.1-Flash-Preview always thinks (`thinking.type: "disabled"` and
+	// effort `none` 400 with error 2013) and tunes depth through
+	// `output_config.effort` from `low` to `max` (the default when omitted).
+	// The identity map replaces the M3 family's adaptive-tag-only map so each
+	// level reaches the wire. Like the M2 family, thinking cannot be turned off:
+	// `thinking-requires-effort` hides the off level and clamps forced-off
+	// requests to the lowest effort instead of sending `disabled`.
+	on "minimax" "minimax-cn" "minimax-code" "minimax-code-cn" {
+		models "MiniMax-M3.1-Flash-Preview" {
+			thinking-requires-effort #true
+			thinking-efforts "low" "medium" "high" "xhigh" "max"
+			thinking-effort-map {
+				low "low"
+				medium "medium"
+				high "high"
+			}
+		}
+	}
+	// Token Plan rows ride the Anthropic route; these OpenAI-only wire axes
+	// serve the legacy lightning rows that stay on the OpenAI-compatible one.
 	on "minimax-code" "minimax-code-cn" {
 		supports-reasoning-effort #false
 		supports-store #false

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

@@ -87,3 +87,105 @@ class "openai" {
 		}
 	}
 }
+
+// GPT-5.2+ bills images in 32px patches × 1.2, sized per `detail` level:
+// https://developers.openai.com/api/docs/guides/images-vision#calculating-costs
+// Older lines carry no lineage rule and bill at their wire API's fallback
+// (providers/image-tokenization.kdl).
+class "openai" {
+	// gpt-5.6 sizing, shared by later lines. `<10` keeps separator-collapsed
+	// revisions (`openai-gpt-52-codex` parses as 52) off the 5.6+ sizing.
+	revision ">=5.6 <10" {
+		image-tokenization {
+			regime "openai-patch"
+			multiplier 1.2
+			auto "original"
+			low {
+				max-edge 512
+			}
+			high {
+				max-edge 2048
+				patch-budget 2500
+			}
+			original {
+				max-edge 65535
+			}
+		}
+	}
+	revision "=5.5" {
+		image-tokenization {
+			regime "openai-patch"
+			multiplier 1.2
+			auto "original"
+			low {
+				max-edge 512
+			}
+			high {
+				max-edge 2048
+				patch-budget 2500
+			}
+			original {
+				max-edge 6000
+				patch-budget 10000
+			}
+		}
+	}
+	// gpt-5.4: `auto` sizes like `high`, and `low` keeps a 6,144-patch budget.
+	revision "=5.4" {
+		image-tokenization {
+			regime "openai-patch"
+			multiplier 1.2
+			auto "high"
+			low {
+				max-edge 2048
+				patch-budget 6144
+			}
+			high {
+				max-edge 2048
+				patch-budget 2500
+			}
+			original {
+				max-edge 6000
+				patch-budget 10000
+			}
+		}
+	}
+	// gpt-5.2 (and the 5.3 Codex SKU) size every level the same way.
+	revision ">=5.2 <5.4" {
+		image-tokenization {
+			regime "openai-patch"
+			multiplier 1.2
+			auto "high"
+			low {
+				max-edge 2048
+				patch-budget 6144
+			}
+			high {
+				max-edge 2048
+				patch-budget 6144
+			}
+			original {
+				max-edge 2048
+				patch-budget 6144
+			}
+		}
+	}
+	// residue: astra's limits differ from the rest of its revision.
+	models "*gpt-6-astra*" {
+		image-tokenization {
+			regime "openai-patch"
+			multiplier 1.2
+			auto "original"
+			low {
+				max-edge 512
+			}
+			high {
+				max-edge 65535
+				patch-budget 2500
+			}
+			original {
+				max-edge 65535
+			}
+		}
+	}
+}

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

@@ -5,4 +5,5 @@
 provider "apple" {
 	default-model "on-device"
 	allow-unauthenticated #true
+	automatic-default #false
 }

package/src/compat/rules/providers/google-antigravity.kdl

@@ -2,6 +2,9 @@
 
 provider "google-antigravity" {
 	default-model "gemini-3.1-pro"
+	// `fetchAvailableModels` is the account's served roster; bundled rows it
+	// omits (e.g. Claude 5.5 on plans without it) 404 on every request (#14328).
+	dynamic-models-authoritative #true
 	kind-apis {
 		image "google-gemini-cli"
 	}

package/src/compat/rules/providers/image-tokenization.kdl

@@ -0,0 +1,42 @@
+// Image billing for readers whose lineage has no `image-tokenization` rule in
+// classes/*.kdl, by the wire API carrying the request. priority=-1: these are
+// fallbacks, so a class-wide lineage rule (same specificity) always wins.
+
+// OpenAI-compatible wires bill like GPT-5.5: 32px patches × 1.2, `auto` sized
+// like `original`.
+on-api "openai-completions" "openai-responses" "openai-codex-responses" "azure-openai-responses" priority=-1 {
+	image-tokenization {
+		regime "openai-patch"
+		multiplier 1.2
+		auto "original"
+		low {
+			max-edge 512
+		}
+		high {
+			max-edge 2048
+			patch-budget 2500
+		}
+		original {
+			max-edge 6000
+			patch-budget 10000
+		}
+	}
+}
+
+// Google wires bill Gemini 3's fixed `media_resolution` budget per image.
+on-api "google-generative-ai" "google-gemini-cli" "google-vertex" priority=-1 {
+	image-tokenization {
+		regime "fixed"
+		tokens 1120
+	}
+}
+
+// Anthropic wires and every other adapter bill Claude's high-res tier, the
+// largest per-image bill, so an unpriced reader is never undercounted.
+on-api "anthropic-messages" "bedrock-converse-stream" "openrouter" "ollama-chat" "cursor-agent" "factory-droid-agent" "gitlab-duo-agent" "devin-agent" "apple-foundation-models" priority=-1 {
+	image-tokenization {
+		regime "anthropic-patch"
+		max-edge 2576
+		max-tokens 4784
+	}
+}

package/src/compat/rules/providers/minimax-code-cn.kdl

@@ -6,11 +6,16 @@ provider "minimax-code-cn" priority=1 {
 
 	// Replaces the provider-id regex in both OpenAI baselines.
 	reasoning-deltas-may-be-cumulative #true
-	class "minimax" {
+	// residue: the legacy lightning SKUs survive only as previous-snapshot rows
+	// on the OpenAI-compatible route (models.dev and both live model lists omit
+	// them), so they keep that route's effort-mode thinking instead of the
+	// Anthropic adaptive rules the rest of the plan uses.
+	models "MiniMax-M2.1-lightning" "MiniMax-M2.5-lightning" {
 		thinking-mode "effort"
-	}
-	// residue: taxonomy ranks and exact globs do not isolate these models.
-	models "MiniMax-M3" {
-		thinking-efforts "minimal" "low" "medium" "high"
+		thinking-effort-map {
+			low "low"
+			medium "medium"
+			high "high"
+		}
 	}
 }

package/src/compat/rules/providers/minimax-code.kdl

@@ -6,11 +6,16 @@ provider "minimax-code" priority=1 {
 
 	// Replaces the provider-id regex in both OpenAI baselines.
 	reasoning-deltas-may-be-cumulative #true
-	class "minimax" {
+	// residue: the legacy lightning SKUs survive only as previous-snapshot rows
+	// on the OpenAI-compatible route (models.dev and both live model lists omit
+	// them), so they keep that route's effort-mode thinking instead of the
+	// Anthropic adaptive rules the rest of the plan uses.
+	models "MiniMax-M2.1-lightning" "MiniMax-M2.5-lightning" {
 		thinking-mode "effort"
-	}
-	// residue: taxonomy ranks and exact globs do not isolate these models.
-	models "MiniMax-M3" {
-		thinking-efforts "minimal" "low" "medium" "high"
+		thinking-effort-map {
+			low "low"
+			medium "medium"
+			high "high"
+		}
 	}
 }

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

@@ -4,6 +4,26 @@ provider "mistral" {
 	default-model "devstral-medium-latest"
 	env "MISTRAL_API_KEY"
 
+	// Public-preview aliases and current USD / 1M token pricing:
+	// https://docs.mistral.ai/models/mistral-large-4-0 (2026-10-06).
+	// Public preview is 50% off: input $0.68, output $2.09, cached input $0.07.
+	// List prices: input $1.36, output $4.18, cached input $0.14; update when preview ends.
+	// No separate maximum output-token limit is published.
+	seed api="openai-completions" base-url="https://api.mistral.ai/v1" bundle="always" {
+		model "mistral-large-4" name="Mistral Large 4" {
+			reasoning #true
+			input "text" "image"
+			cost input=0.68 output=2.09 cache-read=0.07 cache-write=0
+			limits context=1000000
+		}
+		model "mistral-large-4-0" name="Mistral Large 4" {
+			reasoning #true
+			input "text" "image"
+			cost input=0.68 output=2.09 cache-read=0.07 cache-write=0
+			limits context=1000000
+		}
+	}
+
 	thinking-mode "effort"
 	class "unknown" {
 		thinking-efforts "minimal" "low" "medium" "high" "xhigh"

package/src/compat/rules/providers/muse-code.kdl

@@ -15,6 +15,13 @@ provider "muse-code" {
 	// api.meta.ai/v1 accepts only tool_choice "auto"; "none", "required" and named
 	// choices are rejected with 400 (verified 2026-09-10), so omit the field.
 	supports-tool-choice #false
+	// Results can be stored so a dropped stream is recovered: verified 2026-09-29
+	// against api.meta.ai/v1, a stored run keeps generating after the client
+	// socket dies and `GET /responses/{id}` returns it once finished (~60s for
+	// a high-effort turn), while `store: false` runs 404 forever. Storage is
+	// opt-in (privacy: stored runs retain prompts/outputs on Meta's side): omp
+	// setting providers.muse-code.storeResponses or PI_MUSE_STORE_RESPONSES=1.
+	store-responses #true
 	// Subscription tier bills per request; the compact edit prompt cuts the
 	// recurring tool-schema bytes. Verified 2026-09-05 against api.meta.ai/v1:
 	// `custom` tools are rejected with 400 "`custom` tools are not supported

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"
 	}
@@ -75,6 +76,12 @@ provider "openai-codex" {
 		service-tier-cost {
 			flex 0.5
 			priority 2.5
+			// Ultrafast is Astra-only. 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.
+			ultrafast 8
 		}
 		limits-patch {
 			context-window 272000

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

@@ -176,6 +176,11 @@ provider "openai" {
 			cache-read 2.0
 			cache-write 25.0
 		}
+		// Ultrafast is Astra-only: $60/$6/$75/$300 per 1M short-context
+		// (6x the standard card), and the same 6x holds on the long-context tier.
+		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.

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/runtime/behavior.kdl

@@ -351,6 +351,8 @@ 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"