@gajae-code/ai 0.9.3 → 0.9.5

package/CHANGELOG.md

@@ -2,6 +2,21 @@
 
 ## [Unreleased]
 
+## [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": "0.9.3",
+	"version": "0.9.5",
 	"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": "0.9.3",
+		"@gajae-code/utils": "0.9.5",
 		"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 +477,24 @@ 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;
+	}
+	model.contextWindow =
+		model.provider === "openai-codex" || model.api === "openai-codex-responses" ? 272_000 : 1_050_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 +607,9 @@ 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 +742,10 @@ 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/models.ts

@@ -1,6 +1,12 @@
-import { createRequire } from "node:module";
+import { readFileSync } from "node:fs";
 import { isRetiredModelKey } from "./model-retirements";
 import { applyGeneratedModelPolicies, enrichModelThinking } from "./model-thinking";
+// `with { type: "file" }` is embedded by `bun build --compile` and resolves to
+// the bunfs path inside standalone binaries (and to the on-disk path in dev).
+// A plain `createRequire` of a `.json` listed as an extra compile entrypoint is
+// NOT emitted into the bunfs, and its cwd-fallback masks the failure whenever
+// the process runs inside a repo checkout — see PR body for the minimal repro.
+import modelsJsonPath from "./models.json" with { type: "file" };
 import type { Api, KnownProvider, Model, Usage } from "./types";
 import { isClaudeForcedToolChoiceIncapableModelId } from "./utils/tool-choice-capability";
 
@@ -14,16 +20,14 @@ import { isClaudeForcedToolChoiceIncapableModelId } from "./utils/tool-choice-ca
  */
 type BundledCatalog = typeof import("./models.json");
 
-const require = createRequire(import.meta.url);
-const COMPILED_MODELS_PATH = "./packages/ai/src/models.json";
 let bundledCatalog: BundledCatalog | undefined;
 let providerNames: KnownProvider[] | undefined;
 const providerModelRegistry: Map<string, Map<string, Model<Api>>> = new Map();
 
 function getBundledCatalog(): BundledCatalog {
-	bundledCatalog ??= require(
-		process.env.PI_COMPILED === "true" ? COMPILED_MODELS_PATH : "./models.json",
-	) as BundledCatalog;
+	// TS types a .json import as its contents; at runtime `with { type: "file" }`
+	// yields the file path (bunfs path in compiled binaries, disk path in dev).
+	bundledCatalog ??= JSON.parse(readFileSync(modelsJsonPath as unknown as string, "utf8")) as BundledCatalog;
 	return bundledCatalog;
 }
 

package/src/providers/anthropic.ts

@@ -586,15 +586,12 @@ const ANTHROPIC_BUILTIN_TOOL_NAMES = new Set(["web_search", "code_execution", "t
 export const applyClaudeToolPrefix = (name: string, prefixOverride: string = claudeToolPrefix) => {
 	if (!prefixOverride) return name;
 	if (ANTHROPIC_BUILTIN_TOOL_NAMES.has(name.toLowerCase())) return name;
-	const prefix = prefixOverride.toLowerCase();
-	if (name.toLowerCase().startsWith(prefix)) return name;
 	return `${prefixOverride}${name}`;
 };
 
 export const stripClaudeToolPrefix = (name: string, prefixOverride: string = claudeToolPrefix) => {
 	if (!prefixOverride) return name;
-	const prefix = prefixOverride.toLowerCase();
-	if (!name.toLowerCase().startsWith(prefix)) return name;
+	if (!name.startsWith(prefixOverride)) return name;
 	return name.slice(prefixOverride.length);
 };
 
@@ -1325,6 +1322,25 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 				| (ToolCall & { partialJson: string })
 			) & { index: number };
 			const blocks = output.content as Block[];
+			const blocksByAnthropicIndex = new Map<number, Block>();
+			const getBlockByAnthropicIndex = (anthropicIndex: number) => {
+				const block = blocksByAnthropicIndex.get(anthropicIndex);
+				if (!block) return { block: undefined, contentIndex: -1 };
+				return { block, contentIndex: blocks.indexOf(block) };
+			};
+			const trackBlockByAnthropicIndex = (anthropicIndex: number, block: Block) => {
+				// A duplicate start for an active index is a provider-envelope violation;
+				// finalize the orphaned block so no internal stream fields leak into output.
+				const orphaned = blocksByAnthropicIndex.get(anthropicIndex);
+				if (orphaned) {
+					if (orphaned.type === "toolCall" && orphaned.partialJson.trim()) {
+						orphaned.arguments = parseStreamingJson(orphaned.partialJson);
+					}
+					delete (orphaned as { index?: number }).index;
+					delete (orphaned as { partialJson?: string }).partialJson;
+				}
+				blocksByAnthropicIndex.set(anthropicIndex, block);
+			};
 			const idleTimeoutMs = options?.streamIdleTimeoutMs ?? getStreamIdleTimeoutMs();
 			const firstEventTimeoutMs = options?.streamFirstEventTimeoutMs ?? getStreamFirstEventTimeoutMs(idleTimeoutMs);
 			stream.push({ type: "start", partial: output });
@@ -1334,6 +1350,8 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 			let providerRetryAttempt = 0;
 			let thinkingRepairAttempted = false;
 			while (true) {
+				// Retries reset output.content; drop stale block correlations from the aborted attempt.
+				blocksByAnthropicIndex.clear();
 				activeAbortTracker = createAbortSourceTracker(options?.signal);
 				const firstEventTimeoutAbortError = new Error(
 					"Anthropic stream timed out while waiting for the first event",
@@ -1403,6 +1421,7 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 									index: event.index,
 								};
 								output.content.push(block);
+								trackBlockByAnthropicIndex(event.index, block);
 								stream.push({
 									type: "text_start",
 									contentIndex: output.content.length - 1,
@@ -1416,6 +1435,7 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 									index: event.index,
 								};
 								output.content.push(block);
+								trackBlockByAnthropicIndex(event.index, block);
 								stream.push({
 									type: "thinking_start",
 									contentIndex: output.content.length - 1,
@@ -1428,6 +1448,7 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 									index: event.index,
 								};
 								output.content.push(block);
+								trackBlockByAnthropicIndex(event.index, block);
 							} else if (event.content_block.type === "tool_use") {
 								streamedReplayUnsafeContent = true;
 								const block: Block = {
@@ -1441,6 +1462,7 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 									index: event.index,
 								};
 								output.content.push(block);
+								trackBlockByAnthropicIndex(event.index, block);
 								stream.push({
 									type: "toolcall_start",
 									contentIndex: output.content.length - 1,
@@ -1449,8 +1471,7 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 							}
 						} else if (event.type === "content_block_delta") {
 							if (event.delta.type === "text_delta") {
-								const index = blocks.findIndex(b => b.index === event.index);
-								const block = blocks[index];
+								const { block, contentIndex: index } = getBlockByAnthropicIndex(event.index);
 								if (block && block.type === "text") {
 									block.text += event.delta.text;
 									stream.push({
@@ -1461,8 +1482,7 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 									});
 								}
 							} else if (event.delta.type === "thinking_delta") {
-								const index = blocks.findIndex(b => b.index === event.index);
-								const block = blocks[index];
+								const { block, contentIndex: index } = getBlockByAnthropicIndex(event.index);
 								if (block && block.type === "thinking") {
 									block.thinking += event.delta.thinking;
 									stream.push({
@@ -1473,8 +1493,7 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 									});
 								}
 							} else if (event.delta.type === "input_json_delta") {
-								const index = blocks.findIndex(b => b.index === event.index);
-								const block = blocks[index];
+								const { block, contentIndex: index } = getBlockByAnthropicIndex(event.index);
 								if (block && block.type === "toolCall") {
 									block.partialJson += event.delta.partial_json;
 									block.arguments = parseStreamingJson(block.partialJson);
@@ -1486,17 +1505,16 @@ export const streamAnthropic: StreamFunction<"anthropic-messages"> = (
 									});
 								}
 							} else if (event.delta.type === "signature_delta") {
-								const index = blocks.findIndex(b => b.index === event.index);
-								const block = blocks[index];
+								const { block } = getBlockByAnthropicIndex(event.index);
 								if (block && block.type === "thinking") {
 									block.thinkingSignature = block.thinkingSignature || "";
 									block.thinkingSignature += event.delta.signature;
 								}
 							}
 						} else if (event.type === "content_block_stop") {
-							const index = blocks.findIndex(b => b.index === event.index);
-							const block = blocks[index];
+							const { block, contentIndex: index } = getBlockByAnthropicIndex(event.index);
 							if (block) {
+								blocksByAnthropicIndex.delete(event.index);
 								delete (block as { index?: number }).index;
 								if (block.type === "text") {
 									stream.push({
@@ -2245,7 +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,
 		);
 	}