@oh-my-pi/pi-catalog 18.4.2 → 18.4.4

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

@@ -1,13 +1,12 @@
 // Command Code Provider API deployment contract. Model capabilities and effort
-// ladders follow command-code@1.44.0: the thinking-efforts below reproduce the
-// bundle's 36-id effort registry 1:1 (verified against its effort map), plus
-// the 5 Muse Spark lineage ids whose ladder OMP's classes/meta.kdl carries and
-// `deepseek/deepseek-v4.1-flash`, which postdates that bundle (see its rule).
-// Every id the 1.44.0 census serves without an effort dial keeps a missing
-// ladder here as a deliberate absence, not a gap (`modelSupportsReasoning-
-// Effort` is false for those ids in that bundle). Later bundles (1.53.0) hand
-// dials to a few further ids — MiniMaxAI/MiniMax-M3 and meta/muse-spark-1.3's
-// `max` — which stay out of this change.
+// ladders follow command-code@1.67.0. The thinking-efforts below reproduce that
+// bundle's effort registry exactly, including the Muse Spark ids it now lists.
+// An id that 1.67.0 serves without an effort dial has no ladder here on
+// purpose, not by oversight (`modelSupportsReasoningEffort` is false for it in
+// that bundle). The one exception is `claude-sonnet-5-5`: served since
+// 2026-09-29 but not yet in the command-code@1.67.0 registry, so its ladder,
+// image input, and price come from Anthropic's first-party catalog at
+// Anthropic list price.
 provider "commandcode" {
 	default-model "claude-sonnet-5"
 	env "COMMAND_CODE_API_KEY" "COMMANDCODE_API_KEY"
@@ -18,12 +17,24 @@ provider "commandcode" {
 	// backfill reasoning, input, or limits during generation.
 	skip-cross-provider-reference-fills #true
 	discovery label="Command Code" allow-unauthenticated=#true
+	// Command Code serves TypeSafe's jev only through the System One route
+	// (`POST /provider/v1/systemone`); `/provider/v1/models` omits it.
+	seed api="typesafe" base-url="https://api.commandcode.ai/provider" bundle="always" {
+		model "typesafe/jev" name="TypeSafe Jev" {
+			reasoning #false
+			input "text"
+			// Jev bills input only (command-code@1.67.0 display rate after its free window).
+			cost input=0.042 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
+	}
 
 	supports-developer-role #false
 	supports-store #false
-	// No effort dial by default: only the exact ids below expose one
-	// upstream (command-code@1.44.0 effort registry + Muse Spark lineage).
-	// Every other served id must omit the effort control on the wire.
+	// No effort dial by default. Only the exact ids below expose one upstream
+	// (command-code@1.67.0 effort registry); every other served id must omit
+	// the effort control on the wire.
 	supports-reasoning-effort #false
 	// Discovery rows carry no reasoning metadata and stay neutral, so exact
 	// ladders below opt into repairing the neutral default (the cascade
@@ -31,9 +42,10 @@ provider "commandcode" {
 	thinking-upgrade-neutral #true
 	thinking-mode "effort"
 	max-tokens-field "max_tokens"
-	// Default output cap: the Provider API exposes no per-model maximum and
-	// the CLI requests 64K by default, so 64K (2^16); the two registry caps
-	// below are the only per-model maxOutputTokens in command-code@1.44.0.
+	// Default output cap: the Provider API exposes no per-model maximum, and
+	// the CLI requests 64K by default, so the default is 64K (2^16). The
+	// per-model caps below are the only maxOutputTokens that the
+	// command-code@1.67.0 registry sets.
 	limits-patch {
 		max-tokens 65536
 	}
@@ -42,8 +54,13 @@ provider "commandcode" {
 		supports-eager-tool-input-streaming #false
 		supports-long-cache-retention #false
 	}
+	models "typesafe/jev" {
+		kind "judge"
+	}
 	models "claude-fable-5" "claude-fable-5-1" "claude-opus-4-7" "claude-opus-4-8" \
-		"claude-opus-5" "claude-sonnet-4-6" "claude-sonnet-5" {
+		"claude-opus-5" "claude-opus-5-5" "claude-sonnet-4-6" "claude-sonnet-5" "claude-sonnet-5-5" \
+		"gpt-5.6-luna" "gpt-5.6-sol" "gpt-5.6-terra" "gpt-6-astra" "gpt-6-luna" "gpt-6-sol" \
+		"meta/muse-spark-1.3" {
 		supports-reasoning-effort #true
 		thinking-efforts "low" "medium" "high" "xhigh" "max"
 	}
@@ -52,44 +69,27 @@ provider "commandcode" {
 		supports-reasoning-effort #true
 		thinking-efforts "high" "max"
 	}
-	models "deepseek/deepseek-v4-flash-fast" "moonshotai/Kimi-K3" "z-ai/glm-5.3-flash" \
-		"zai-org/GLM-5.3" {
-		supports-reasoning-effort #true
-		thinking-efforts "low" "high" "max"
-	}
-	// `deepseek/deepseek-v4.1-flash` postdates command-code@1.44.0, so it is
-	// absent from that bundle's effort registry; command-code@1.53.0 lists it
-	// in the same low/high/max group as Kimi K3 and GLM-5.3, and the live docs
-	// model table carries the same scale plus image input.
-	models "deepseek/deepseek-v4.1-flash" {
+	models "deepseek/deepseek-v4-flash-fast" "deepseek/deepseek-v4.1-flash" \
+		"deepseek/deepseek-v4.1-flash-fast" "moonshotai/Kimi-K3" "z-ai/glm-5.3-flash" \
+		"z-ai/glm-5.3-flashx" "zai-org/GLM-5.3" {
 		supports-reasoning-effort #true
 		thinking-efforts "low" "high" "max"
-		// The class's `token="vision"` exemption misses this id (its name
-		// carries no `vision` token) even though the deployment serves it with
-		// image input, so the opt-out is explicit here.
-		strip-image-input #false
-		input-modalities "text" "image"
 	}
 	models "google/gemini-3.1-flash-lite" "google/gemini-3.5-flash" \
 		"google/gemini-3.5-flash-lite" "google/gemini-3.6-flash" "google/gemini-3.7-flash" \
-		"google/gemini-3.8-flash" "gpt-5.4-mini" "xai/grok-4.5" {
+		"google/gemini-3.8-flash" "gpt-5.4-mini" "MiniMaxAI/MiniMax-M3" \
+		"stealth/space-bunny-alpha" "stepfun/Step-5-Preview" "tencent/hy4-preview" "xai/grok-4.5" {
 		supports-reasoning-effort #true
 		thinking-efforts "low" "medium" "high"
 	}
-	models "gpt-5.3-codex" "gpt-5.4" "gpt-5.5" "xai/grok-4.6" {
+	models "gpt-5.3-codex" "gpt-5.4" "gpt-5.5" "meta/muse-spark-1.1" "meta/muse-spark-1.2" \
+		"meta/muse-spark-1.2-contributor" "meta/muse-spark-1.3-contributor" "xai/grok-4.6" \
+		"xai/grok-4.7" {
 		supports-reasoning-effort #true
 		thinking-efforts "low" "medium" "high" "xhigh"
 	}
-	models "gpt-5.6-luna" "gpt-5.6-sol" "gpt-5.6-terra" {
-		supports-reasoning-effort #true
-		thinking-efforts "low" "medium" "high" "xhigh" "max"
-	}
-	models "meta/muse-spark-1.1" "meta/muse-spark-1.2" "meta/muse-spark-1.2-contributor" \
-		"meta/muse-spark-1.3" "meta/muse-spark-1.3-contributor" {
-		supports-reasoning-effort #true
-		thinking-efforts "minimal" "low" "medium" "high" "xhigh"
-	}
-	models "Qwen/Qwen3.8-27B" "Qwen/Qwen3.8-Flash" "Qwen/Qwen3.8-Max" "Qwen/Qwen3.8-Max-0902" {
+	models "Qwen/Qwen3.8-27B" "Qwen/Qwen3.8-Flash" "Qwen/Qwen3.8-Max" "Qwen/Qwen3.8-Max-0902" \
+		"Qwen/Qwen3.8-Omni-Flash" "stealth/pixel-canary" {
 		supports-reasoning-effort #true
 		thinking-efforts "low" "medium" "xhigh"
 	}
@@ -97,20 +97,54 @@ provider "commandcode" {
 		supports-reasoning-effort #true
 		thinking-efforts "high" "xhigh"
 	}
-	models "tencent/hy4-preview" {
-		supports-reasoning-effort #true
-		thinking-efforts "low" "medium" "high"
+	// Image input follows command-code@1.67.0: every served id outside its text-only set.
+	// residue: the vendor's text-only set cuts across classes (Qwen, DeepSeek, GLM,
+	// MiniMax, and Xiaomi each have ids on both sides), so no class selector
+	// expresses this list.
+	models "claude-fable-5" "claude-fable-5-1" "claude-haiku-4-5-20251001" "claude-opus-4-7" \
+		"claude-opus-4-8" "claude-opus-5" "claude-opus-5-5" "claude-sonnet-4-6" "claude-sonnet-5" \
+		"claude-sonnet-5-5" \
+		"deepseek/deepseek-v4-flash-vision-exp" "deepseek/deepseek-v4.1-flash-fast" \
+		"google/gemini-3.1-flash-lite" \
+		"google/gemini-3.5-flash" "google/gemini-3.5-flash-lite" "google/gemini-3.6-flash" \
+		"google/gemini-3.7-flash" "google/gemini-3.8-flash" "gpt-5.3-codex" "gpt-5.4" \
+		"gpt-5.4-mini" "gpt-5.5" "gpt-5.6-luna" "gpt-5.6-sol" "gpt-5.6-terra" "gpt-6-astra" \
+		"gpt-6-luna" "gpt-6-sol" "meta/muse-spark-1.1" "meta/muse-spark-1.2" \
+		"meta/muse-spark-1.2-contributor" "meta/muse-spark-1.3" "meta/muse-spark-1.3-contributor" \
+		"MiniMaxAI/MiniMax-M3" "moonshotai/Kimi-K2.5" "moonshotai/Kimi-K2.6" \
+		"moonshotai/Kimi-K2.7-Code" "moonshotai/Kimi-K2.7-Code-Highspeed" "moonshotai/Kimi-K3" \
+		"Qwen/Qwen3.6-Plus" "Qwen/Qwen3.7-Flash" "Qwen/Qwen3.7-Plus" "Qwen/Qwen3.8-27B" \
+		"Qwen/Qwen3.8-Flash" "Qwen/Qwen3.8-Max" "Qwen/Qwen3.8-Max-0902" "Qwen/Qwen3.8-Omni-Flash" \
+		"sakana/fugu-ultra" "stealth/pixel-canary" "stealth/space-bunny-alpha" \
+		"stepfun/Step-3.7-Flash" "stepfun/Step-5-Preview" "thinkingmachines/inkling" \
+		"thinkingmachines/inkling-small" "xai/grok-4.5" "xai/grok-4.6" "xai/grok-4.7" \
+		"xiaomi/mimo-v2.5" "xiaomi/mimo-v2.6-flash" "xiaomi/mimo-v2.6-pro" \
+		"xiaomi/mimo-v2.6-pro-ultraspeed" "z-ai/glm-5.3-flash" "z-ai/glm-5.3-flashx" {
+		input-modalities "text" "image"
 	}
-	models "Qwen/Qwen3.8-27B" {
+	// The class's `token="vision"` exemption misses this id (its name carries no
+	// `vision` token) even though the deployment serves it with image input, so
+	// the opt-out is explicit here.
+	models "deepseek/deepseek-v4.1-flash" {
+		strip-image-input #false
+		input-modalities "text" "image"
+	}
+	models "inclusionai/ling-3.0-flash-sante:free" "poolside/laguna-s-2.1-free" "Qwen/Qwen3.8-27B" {
 		limits-patch {
 			max-tokens 32768
 		}
 	}
-	models "z-ai/glm-5.3-flash" {
+	models "Qwen/Qwen3.8-Omni-Flash" "stealth/pixel-canary" "z-ai/glm-5.3-flash" \
+		"z-ai/glm-5.3-flashx" {
 		limits-patch {
 			max-tokens 131072
 		}
 	}
+	models "stealth/space-bunny-alpha" {
+		limits-patch {
+			max-tokens 524288
+		}
+	}
 	// Codex SKUs report the output budget inside the advertised window, so
 	// the class lineage pins the 272K prompt window; restate it here or the
 	// class rule wins the whole limits axis and drops the 64K provider cap.
@@ -120,22 +154,32 @@ provider "commandcode" {
 			max-tokens 65536
 		}
 	}
+	// Command Code's Responses endpoint rejects an object `tool_choice` and returns
+	// no bytes from `image_generation` (verified 2026-09-29), so the openai
+	// class's hosted-image and image-model grants do not apply and named choices
+	// degrade to `"required"`. Hosted web search still works with a string `tool_choice`.
+	models "gpt-5.3-codex" "gpt-5.4" "gpt-5.4-mini" "gpt-5.5" "gpt-5.6-luna" "gpt-5.6-sol" \
+		"gpt-5.6-terra" "gpt-6-astra" "gpt-6-luna" "gpt-6-sol" {
+		hosted-image #false
+		image-model #false
+		supports-named-tool-choice #false
+	}
 	// ---------------------------------------------------------------------------
-	// Provider pricing. Every model id served by GET /provider/v1/models is
-	// priced here; discovery defaults each model to zero cost and the mapper
-	// never inherits another provider's rates, so this table is the sole
-	// pricing source. Base rates are Command Code's per-1M-token USD list
-	// prices, cross-checked 1:1 against the rate card bundled in
-	// command-code@1.44.0 (67 rows, zero mismatches) and the live catalog on
-	// 2026-09-09 (68 rows: the bundle plus `inclusionai/ling-3.0-flash-sante:free`).
-	// The page's Sonnet 4.5 / Opus 4.6 / GPT-6 Astra rows are not served by
-	// the Provider API and stay absent. A "—" Cache Write column on the page
-	// means 0. Standing deals (MiniMax M3 50% off, MiMo V2.5 discounts) are
-	// priced at their current effective rates. Free models
-	// (`poolside/laguna-s-2.1-free`, `meituan/LongCat-2.0:free`,
-	// `inclusionai/ling-3.0-flash-sante:free`) keep the zero discovery default.
+	// This table prices every model id that GET /provider/v1/models serves.
+	// Discovery defaults each model to zero cost and the mapper never inherits
+	// another provider's rates, so this table is the only pricing source.
+	// Base rates are Command Code's per-1M-token USD display rates. They follow
+	// `getDisplayRates` in command-code@1.67.0, read on 2026-09-29. That table
+	// prices 83 of the 84 ids the live catalog served that day; it has no
+	// `claude-sonnet-5-5`. DeepSeek V4 Flash and its vision-exp variant bill on
+	// the V4.1 Flash gateway slug.
+	// A missing cache-write rate means 0. Standing deals (MiniMax M3 50% off,
+	// MiMo discounts) use their current effective rates.
+	// Free models (`poolside/laguna-s-2.1-free`,
+	// `inclusionai/ling-3.0-flash-sante:free`, `stealth/pixel-canary`,
+	// `stealth/space-bunny-alpha`) keep the zero discovery default.
 	// ---------------------------------------------------------------------------
-	models "claude-sonnet-5" {
+	models "claude-sonnet-5" "claude-sonnet-5-5" {
 		cost-patch {
 			input 2.0
 			output 10.0
@@ -176,6 +220,14 @@ provider "commandcode" {
 			cache-write 6.25
 		}
 	}
+	models "claude-opus-5-5" {
+		cost-patch {
+			input 4.0
+			output 20.0
+			cache-read 0.2
+			cache-write 5.0
+		}
+	}
 	models "claude-haiku-4-5-20251001" {
 		cost-patch {
 			input 1.0
@@ -184,6 +236,51 @@ provider "commandcode" {
 			cache-write 1.25
 		}
 	}
+	models "gpt-6-astra" {
+		cost-patch {
+			input 10.0
+			output 50.0
+			cache-read 1.0
+			cache-write 12.5
+		}
+		long-context-cost {
+			input-threshold 272000
+			input 20.0
+			output 75.0
+			cache-read 2.0
+			cache-write 25.0
+		}
+	}
+	models "gpt-6-sol" {
+		cost-patch {
+			input 2.0
+			output 10.0
+			cache-read 0.2
+			cache-write 2.5
+		}
+		long-context-cost {
+			input-threshold 272000
+			input 4.0
+			output 15.0
+			cache-read 0.4
+			cache-write 5.0
+		}
+	}
+	models "gpt-6-luna" {
+		cost-patch {
+			input 0.1
+			output 0.5
+			cache-read 0.01
+			cache-write 0.125
+		}
+		long-context-cost {
+			input-threshold 272000
+			input 0.2
+			output 0.75
+			cache-read 0.02
+			cache-write 0.25
+		}
+	}
 	models "gpt-5.6-sol" {
 		cost-patch {
 			input 5.0
@@ -243,6 +340,7 @@ provider "commandcode" {
 	// DeepSeek V4 bills time-dependent rates; the page displays the off-peak
 	// rates that apply 17h/day (peak doubles input/output). Off-peak is the
 	// documented display convention; the Usage page stays authoritative.
+	// V4.1 Flash Fast displays 0.16 / 0.58 / 0.016.
 	models "deepseek/deepseek-v4-pro" {
 		cost-patch {
 			input 0.66
@@ -251,11 +349,14 @@ provider "commandcode" {
 			cache-write 0.0
 		}
 	}
-	models "deepseek/deepseek-v4-flash" "deepseek/deepseek-v4-flash-vision-exp" {
+	// command-code@1.67.0 bills the flash id and its vision-exp variant on the
+	// V4.1 Flash gateway slug, so all three ids share the V4.1 Flash off-peak rates.
+	models "deepseek/deepseek-v4-flash" "deepseek/deepseek-v4-flash-vision-exp" \
+		"deepseek/deepseek-v4.1-flash" {
 		cost-patch {
-			input 0.22
-			output 0.66
-			cache-read 0.007
+			input 0.15
+			output 0.6
+			cache-read 0.003
 			cache-write 0.0
 		}
 	}
@@ -267,13 +368,11 @@ provider "commandcode" {
 			cache-write 0.0
 		}
 	}
-	// Served by the live catalog but absent from command-code@1.44.0: models
-	// page off-peak rates as listed 2026-09-10 (cache-write "—" means 0).
-	models "deepseek/deepseek-v4.1-flash" {
+	models "deepseek/deepseek-v4.1-flash-fast" {
 		cost-patch {
-			input 0.15
-			output 0.6
-			cache-read 0.003
+			input 0.16
+			output 0.58
+			cache-read 0.016
 			cache-write 0.0
 		}
 	}
@@ -327,6 +426,14 @@ provider "commandcode" {
 			cache-write 0.0
 		}
 	}
+	models "z-ai/glm-5.3-flashx" {
+		cost-patch {
+			input 0.37
+			output 1.25
+			cache-read 0.075
+			cache-write 0.0
+		}
+	}
 	models "zai-org/GLM-5.3" "zai-org/GLM-5.2" "zai-org/GLM-5.1" {
 		cost-patch {
 			input 1.4
@@ -368,9 +475,17 @@ provider "commandcode" {
 			cache-write 0.0
 		}
 	}
-	// MiMo V2.5 carries standing Xiaomi discounts; priced at the current
+	models "meituan/LongCat-2.0" {
+		cost-patch {
+			input 0.3
+			output 1.2
+			cache-read 0.006
+			cache-write 0.0
+		}
+	}
+	// MiMo carries standing Xiaomi discounts; priced at the current
 	// unified-context effective rates.
-	models "xiaomi/mimo-v2.5-pro" {
+	models "xiaomi/mimo-v2.5-pro" "xiaomi/mimo-v2.6-pro" {
 		cost-patch {
 			input 0.435
 			output 0.87
@@ -378,7 +493,7 @@ provider "commandcode" {
 			cache-write 0.0
 		}
 	}
-	models "xiaomi/mimo-v2.5" {
+	models "xiaomi/mimo-v2.5" "xiaomi/mimo-v2.6-flash" {
 		cost-patch {
 			input 0.14
 			output 0.28
@@ -386,6 +501,14 @@ provider "commandcode" {
 			cache-write 0.0
 		}
 	}
+	models "xiaomi/mimo-v2.6-pro-ultraspeed" {
+		cost-patch {
+			input 4.35
+			output 8.7
+			cache-read 0.036
+			cache-write 0.0
+		}
+	}
 	models "Qwen/Qwen3.8-Max-0902" {
 		cost-patch {
 			input 2.0
@@ -418,6 +541,14 @@ provider "commandcode" {
 			cache-write 0.0
 		}
 	}
+	models "Qwen/Qwen3.8-Omni-Flash" {
+		cost-patch {
+			input 0.15
+			output 0.47
+			cache-read 0.016
+			cache-write 0.0
+		}
+	}
 	models "Qwen/Qwen3.7-Max" {
 		cost-patch {
 			input 2.5
@@ -486,12 +617,20 @@ provider "commandcode" {
 	}
 	models "stepfun/Step-3.5-Flash" {
 		cost-patch {
-			input 0.1
+			input 0.09
 			output 0.3
 			cache-read 0.02
 			cache-write 0.0
 		}
 	}
+	models "stepfun/Step-5-Preview" {
+		cost-patch {
+			input 1.0
+			output 2.7
+			cache-read 0.05
+			cache-write 0.0
+		}
+	}
 	models "tencent/hy4-preview" {
 		cost-patch {
 			input 0.834
@@ -624,4 +763,12 @@ provider "commandcode" {
 			cache-write 0.0
 		}
 	}
+	models "xai/grok-4.7" {
+		cost-patch {
+			input 1.2
+			output 3.6
+			cache-read 0.3
+			cache-write 0.0
+		}
+	}
 }

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

@@ -9,7 +9,8 @@ provider "openai-codex" {
 	web-search "codex"
 
 	seed api="openai-codex-responses" base-url="https://chatgpt.com/backend-api" bundle="always" {
-		model "gpt-image-1" name="GPT Image 1" {
+		// Label only: the Codex backend ignores `tool.model` and runs gpt-image-2-codex.
+		model "gpt-image-2" name="GPT Image 2" {
 			reasoning #false
 			input "text" "image"
 			cost input=0 output=0 cache-read=0 cache-write=0
@@ -18,7 +19,7 @@ provider "openai-codex" {
 		}
 	}
 
-	models "gpt-image-1" {
+	models "gpt-image-2" {
 		kind "image"
 	}
 
@@ -88,6 +89,23 @@ provider "openai-codex" {
 			cache-write 0
 		}
 	}
+	// GPT-6.1 Sol keeps GPT-6 Sol's $2/$10 list price but halves cached input
+	// to $0.10 (OpenAI API docs, 2026-09-29). Like Astra, Codex reports a 272K
+	// default and an 872K maximum while OpenAI documents 922K max input (1.05M
+	// total); the curated maximum applies only with extended context on, and a
+	// higher live maximum still wins.
+	models "gpt-6.1-sol" "gpt-6.1-sol-wm" {
+		cost-patch {
+			input 2.0
+			output 10.0
+			cache-read 0.1
+			cache-write 0
+		}
+		limits-patch {
+			context-window 272000
+		}
+		max-context-window 922000
+	}
 	models "gpt-6-luna" "gpt-6-luna-wm" {
 		cost-patch {
 			input 0.1

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

@@ -3,12 +3,6 @@
 provider "openai" {
 	default-model "gpt-5.5"
 	env "OPENAI_API_KEY"
-	// Responses web search is documented for GPT-5.5 and GPT-6 Astra, and has
-	// been live-verified for GPT-6 Luna and GPT-5.6 Luna. Keep this capability
-	// model-scoped so realtime and other unverified OpenAI models stay excluded.
-	models "gpt-5.5" "gpt-6-astra" "gpt-6-luna" "gpt-5.6-luna" {
-		web-search "openai"
-	}
 	kind-apis {
 		embedding "openai-embeddings"
 		image "openai-responses"

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

@@ -5,6 +5,14 @@ provider "web" {
 	allow-unauthenticated #true
 
 	seed api="web-search" base-url="web://search" bundle="always" {
+		// Placeholder resolved at search time to the session's active model, when it declares a grounding.
+		model "hosted" name="Session model" {
+			reasoning #false
+			input "text"
+			cost input=0 output=0 cache-read=0 cache-write=0
+			limits
+			supports-tools #false
+		}
 		model "parallel" name="Parallel" {
 			reasoning #false
 			input "text"

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

@@ -191,10 +191,15 @@ behavior {
 			glob="claude-fable-*" glob="claude-mythos-*"
 		route "openai-responses" prefix="grok-4." prefix="gpt-5" prefix="gpt-6" prefix="oswe" prefix="mai-"
 	}
-	// Command Code serves Claude models through its Anthropic-compatible
-	// Messages endpoint and every other model through chat completions.
+	// Command Code serves Claude ids through its Anthropic-compatible Messages
+	// endpoint, its ten GPT ids (the same set command-code@1.67.0 marks
+	// `spec: "responses"`) through its Responses endpoint, and every other id
+	// through chat completions.
 	api-routes provider="commandcode" default="openai-completions" {
 		route "anthropic-messages" prefix="claude-"
+		route "openai-responses" exact="gpt-5.3-codex" exact="gpt-5.4" exact="gpt-5.4-mini" exact="gpt-5.5" \
+			exact="gpt-5.6-luna" exact="gpt-5.6-sol" exact="gpt-5.6-terra" exact="gpt-6-astra" \
+			exact="gpt-6-luna" exact="gpt-6-sol"
 	}
 	// ZenMux fronts Anthropic ids on a dedicated Messages endpoint; every
 	// other namespace stays on chat completions.

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

@@ -636,6 +636,9 @@ collapse {
 	provider-alias "devin" "grok-4.6" "grok-4-6"
 	provider-alias "devin" "swe-1.7" "swe-1-7"
 	provider-alias "devin" "swe-1.7-lightning" "swe-1-7-lightning"
+	// 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"
 	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"