@gajae-code/ai 0.9.4 → 0.9.6

package/CHANGELOG.md

@@ -2,10 +2,25 @@
 
 ## [Unreleased]
 
+## [0.9.6] - 2026-07-10
+### Fixed
+
+- Normalized the GPT-5.6 Sol/Terra/Luna context window to the 373K usable prompt budget on both OpenAI and OpenAI code transports (was 1,050K / 272K), matching the live openai-codex catalog.
+
+## [0.9.5] - 2026-07-09
+### Added
+
+- Added GPT-5.6 Sol, Terra, and Luna catalog/parser support for OpenAI and OpenAI code transports, including `low` through canonical `max` reasoning efforts, verified pricing/limits, and GPT-5.6 cache-write pricing (#1925; OmX #3103).
+
+### Fixed
+
+- Stopped requesting `strict: true` tool use on Anthropic OAuth requests: the Claude Code OAuth surface mishandles strict tools, returning tool calls with empty/undefined arguments and occasionally corrupted tool names. API-key requests keep strict tool use; `PI_NO_STRICT=1` is no longer needed as a workaround.
+
 ## [0.9.4] - 2026-07-09
 ### Fixed
 
 - Preserved Anthropic OAuth tool-call names and streamed arguments across interleaved tool-use blocks, preventing prefixed tool names and partial JSON deltas from being dropped or misattributed.
+- Embedded `models.json` via a `with { type: "file" }` import so compiled release binaries load the bundled model catalog from bunfs instead of crashing at startup with `Cannot find module './packages/ai/src/models.json'` (v0.9.3 regression, #1914).
 
 ## [0.9.2] - 2026-07-09
 ### Added

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@gajae-code/ai",
-	"version": "0.9.4",
+	"version": "0.9.6",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"homepage": "https://gajae-code.com",
 	"author": "Yeachan-Heo and Gajae Code Contributors",
@@ -40,7 +40,7 @@
 	"dependencies": {
 		"@anthropic-ai/sdk": "^0.94.0",
 		"@bufbuild/protobuf": "^2.12.0",
-		"@gajae-code/utils": "0.9.4",
+		"@gajae-code/utils": "0.9.6",
 		"openai": "^6.36.0",
 		"partial-json": "^0.1.7",
 		"zod": "4.4.3"

package/src/model-thinking.ts

@@ -47,6 +47,7 @@ const DEFAULT_REASONING_EFFORTS_WITH_XHIGH_AND_MAX: readonly Effort[] = [
 const GEMINI_3_PRO_EFFORTS: readonly Effort[] = [Effort.Low, Effort.High];
 const GEMINI_3_FLASH_EFFORTS: readonly Effort[] = [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High];
 const GPT_5_2_PLUS_EFFORTS: readonly Effort[] = [Effort.Low, Effort.Medium, Effort.High, Effort.XHigh];
+const GPT_5_6_PLUS_EFFORTS: readonly Effort[] = [Effort.Low, Effort.Medium, Effort.High, Effort.XHigh, Effort.Max];
 const GPT_5_5_DEFAULT_EFFORT = Effort.XHigh;
 
 const GPT_5_1_CODEX_MINI_EFFORTS: readonly Effort[] = [Effort.Medium, Effort.High];
@@ -60,7 +61,18 @@ type SemVer = {
 
 type GeminiKind = "pro" | "flash";
 type AnthropicKind = "opus" | "sonnet";
-type OpenAIVariant = "base" | "codex" | "codex-max" | "codex-mini" | "codex-spark" | "mini" | "max" | "nano";
+type OpenAIVariant =
+	| "base"
+	| "codex"
+	| "codex-max"
+	| "codex-mini"
+	| "codex-spark"
+	| "luna"
+	| "mini"
+	| "max"
+	| "nano"
+	| "sol"
+	| "terra";
 
 const CODEX_GPT_5_4_PRIORITY_BY_VARIANT: Partial<Record<OpenAIVariant, number>> = {
 	base: 0,
@@ -465,11 +477,25 @@ function applyGpt55ContextWindow(model: ApiModel<Api>, parsedModel: OpenAIModel)
 	}
 	return false;
 }
+const GPT_5_6_TIER_VARIANTS: ReadonlySet<OpenAIVariant> = new Set(["base", "sol", "terra", "luna"]);
+
+function applyGpt56ContextWindow(model: ApiModel<Api>, parsedModel: OpenAIModel): boolean {
+	if (!semverGte(parsedModel.version, "5.6") || !GPT_5_6_TIER_VARIANTS.has(parsedModel.variant)) {
+		return false;
+	}
+	// GPT-5.6 tiers enforce a ~373K usable prompt budget on both transports
+	// (matches the openai-codex live catalog), despite the 1M+ marketing window.
+	model.contextWindow = 373_000;
+	return true;
+}
 
 function applyOpenAICatalogPolicy(model: ApiModel<Api>, parsedModel: OpenAIModel): void {
 	if (applyGpt55ContextWindow(model, parsedModel)) {
 		return;
 	}
+	if (applyGpt56ContextWindow(model, parsedModel)) {
+		return;
+	}
 	// OpenAI code backend models: 400K figure includes output budget; input window is 272K.
 	if (parsedModel.variant.startsWith("codex") && parsedModel.variant !== "codex-spark") {
 		model.contextWindow = 272000;
@@ -582,6 +608,9 @@ function inferOpenAISupportedEfforts(model: OpenAIModel): readonly Effort[] {
 	if (model.variant === "codex-mini" && semverEqual(model.version, "5.1")) {
 		return GPT_5_1_CODEX_MINI_EFFORTS;
 	}
+	if (semverGte(model.version, "5.6")) {
+		return GPT_5_6_PLUS_EFFORTS;
+	}
 	if (semverGte(model.version, "5.2")) {
 		return GPT_5_2_PLUS_EFFORTS;
 	}
@@ -714,7 +743,10 @@ function parseAnthropicModel(modelId: string): AnthropicModel | null {
 }
 
 function parseOpenAIModel(modelId: string): OpenAIModel | null {
-	const match = /gpt-(\d+(?:\.\d+){0,2})(?:-(codex-spark|codex-mini|codex-max|codex|mini|max|nano))?$/.exec(modelId);
+	const match =
+		/gpt-(\d+(?:\.\d+){0,2})(?:-(codex-spark|codex-mini|codex-max|codex|luna|mini|max|nano|sol|terra))?$/.exec(
+			modelId,
+		);
 	if (!match) {
 		return null;
 	}

package/src/models.json

@@ -56291,6 +56291,84 @@
 				"maxLevel": "xhigh"
 			}
 		},
+		"gpt-5.6-luna": {
+			"id": "gpt-5.6-luna",
+			"name": "GPT-5.6 Luna",
+			"api": "openai-responses",
+			"provider": "openai",
+			"baseUrl": "",
+			"reasoning": true,
+			"input": [
+				"text",
+				"image"
+			],
+			"cost": {
+				"input": 1,
+				"output": 6,
+				"cacheRead": 0.1,
+				"cacheWrite": 1.25
+			},
+			"contextWindow": 1050000,
+			"maxTokens": 128000,
+			"thinking": {
+				"mode": "effort",
+				"minLevel": "low",
+				"maxLevel": "max"
+			},
+			"applyPatchToolType": "freeform"
+		},
+		"gpt-5.6-sol": {
+			"id": "gpt-5.6-sol",
+			"name": "GPT-5.6 Sol",
+			"api": "openai-responses",
+			"provider": "openai",
+			"baseUrl": "",
+			"reasoning": true,
+			"input": [
+				"text",
+				"image"
+			],
+			"cost": {
+				"input": 5,
+				"output": 30,
+				"cacheRead": 0.5,
+				"cacheWrite": 6.25
+			},
+			"contextWindow": 1050000,
+			"maxTokens": 128000,
+			"thinking": {
+				"mode": "effort",
+				"minLevel": "low",
+				"maxLevel": "max"
+			},
+			"applyPatchToolType": "freeform"
+		},
+		"gpt-5.6-terra": {
+			"id": "gpt-5.6-terra",
+			"name": "GPT-5.6 Terra",
+			"api": "openai-responses",
+			"provider": "openai",
+			"baseUrl": "",
+			"reasoning": true,
+			"input": [
+				"text",
+				"image"
+			],
+			"cost": {
+				"input": 2.5,
+				"output": 15,
+				"cacheRead": 0.25,
+				"cacheWrite": 3.125
+			},
+			"contextWindow": 1050000,
+			"maxTokens": 128000,
+			"thinking": {
+				"mode": "effort",
+				"minLevel": "low",
+				"maxLevel": "max"
+			},
+			"applyPatchToolType": "freeform"
+		},
 		"o1": {
 			"id": "o1",
 			"name": "o1",
@@ -56938,6 +57016,87 @@
 				"defaultLevel": "xhigh"
 			},
 			"applyPatchToolType": "freeform"
+		},
+		"gpt-5.6-luna": {
+			"id": "gpt-5.6-luna",
+			"name": "GPT-5.6 Luna",
+			"api": "openai-codex-responses",
+			"provider": "openai-codex",
+			"baseUrl": "https://chatgpt.com/backend-api",
+			"reasoning": true,
+			"input": [
+				"text",
+				"image"
+			],
+			"cost": {
+				"input": 1,
+				"output": 6,
+				"cacheRead": 0.1,
+				"cacheWrite": 1.25
+			},
+			"contextWindow": 272000,
+			"maxTokens": 128000,
+			"preferWebsockets": true,
+			"thinking": {
+				"mode": "effort",
+				"minLevel": "low",
+				"maxLevel": "max"
+			},
+			"applyPatchToolType": "freeform"
+		},
+		"gpt-5.6-sol": {
+			"id": "gpt-5.6-sol",
+			"name": "GPT-5.6 Sol",
+			"api": "openai-codex-responses",
+			"provider": "openai-codex",
+			"baseUrl": "https://chatgpt.com/backend-api",
+			"reasoning": true,
+			"input": [
+				"text",
+				"image"
+			],
+			"cost": {
+				"input": 5,
+				"output": 30,
+				"cacheRead": 0.5,
+				"cacheWrite": 6.25
+			},
+			"contextWindow": 272000,
+			"maxTokens": 128000,
+			"preferWebsockets": true,
+			"thinking": {
+				"mode": "effort",
+				"minLevel": "low",
+				"maxLevel": "max"
+			},
+			"applyPatchToolType": "freeform"
+		},
+		"gpt-5.6-terra": {
+			"id": "gpt-5.6-terra",
+			"name": "GPT-5.6 Terra",
+			"api": "openai-codex-responses",
+			"provider": "openai-codex",
+			"baseUrl": "https://chatgpt.com/backend-api",
+			"reasoning": true,
+			"input": [
+				"text",
+				"image"
+			],
+			"cost": {
+				"input": 2.5,
+				"output": 15,
+				"cacheRead": 0.25,
+				"cacheWrite": 3.125
+			},
+			"contextWindow": 272000,
+			"maxTokens": 128000,
+			"preferWebsockets": true,
+			"thinking": {
+				"mode": "effort",
+				"minLevel": "low",
+				"maxLevel": "max"
+			},
+			"applyPatchToolType": "freeform"
 		}
 	},
 	"opencode": {

package/src/providers/anthropic.ts

@@ -2263,7 +2263,11 @@ function buildParams(
 		params.tools = convertTools(
 			context.tools,
 			isOAuthToken,
-			disableStrictTools || model.provider === "github-copilot",
+			// The Claude Code OAuth surface mishandles `strict: true` tools:
+			// streamed tool_use blocks arrive with empty/undefined arguments and
+			// occasionally corrupted names (works with PI_NO_STRICT=1). Never
+			// request strict tool use on OAuth requests.
+			disableStrictTools || isOAuthToken || model.provider === "github-copilot",
 			getAnthropicCompat(model).supportsEagerToolInputStreaming,
 		);
 	}