@oh-my-pi/pi-ai 13.13.2 → 13.14.2

package/CHANGELOG.md

@@ -2,6 +2,30 @@
 
 ## [Unreleased]
 
+## [13.14.2] - 2026-03-21
+### Changed
+
+- Updated thinking configuration format from `levels` array to `minLevel` and `maxLevel` properties for improved clarity
+- Corrected context window from 400000 to 272000 tokens for GPT-5.4 mini and nano variants on Codex transport
+- Normalized GPT-5.4 variant priority handling to use parsed variant instead of special-casing raw model IDs
+- Added support for `mini` variant in OpenAI model parsing regex
+
+### Fixed
+
+- Fixed inconsistent thinking level configuration across multiple model definitions
+
+## [13.14.0] - 2026-03-20
+
+### Fixed
+
+- Fixed resumed OpenAI Responses sessions to avoid replaying stale same-provider native history on the first follow-up after process restart ([#488](https://github.com/can1357/oh-my-pi/issues/488))
+
+### Added
+
+- Added bundled GPT-5.4 mini model metadata for OpenAI, OpenAI Codex, and GitHub Copilot, including low-to-xhigh thinking support and GitHub Copilot premium multiplier metadata
+- Added bundled GPT-5.4 nano model metadata for OpenAI and OpenAI Codex, including low-to-xhigh thinking support
+
+
 ## [13.13.2] - 2026-03-18
 ### Changed
 

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-ai",
-	"version": "13.13.2",
+	"version": "13.14.2",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"homepage": "https://github.com/can1357/oh-my-pi",
 	"author": "Can Boluk",
@@ -41,7 +41,7 @@
 		"@aws-sdk/client-bedrock-runtime": "^3",
 		"@bufbuild/protobuf": "^2.11",
 		"@google/genai": "^1.43",
-		"@oh-my-pi/pi-utils": "13.13.2",
+		"@oh-my-pi/pi-utils": "13.14.2",
 		"@sinclair/typebox": "^0.34",
 		"@smithy/node-http-handler": "^4.4",
 		"ajv": "^8.18",

package/src/model-thinking.ts

@@ -40,7 +40,13 @@ type SemVer = {
 
 type GeminiKind = "pro" | "flash";
 type AnthropicKind = "opus" | "sonnet";
-type OpenAIVariant = "base" | "codex" | "codex-max" | "codex-mini" | "codex-spark" | "max" | "nano";
+type OpenAIVariant = "base" | "codex" | "codex-max" | "codex-mini" | "codex-spark" | "mini" | "max" | "nano";
+
+const CODEX_GPT_5_4_PRIORITY_BY_VARIANT: Partial<Record<OpenAIVariant, number>> = {
+	base: 0,
+	mini: 1,
+	nano: 2,
+};
 
 interface GeminiModel {
 	family: "gemini";
@@ -296,6 +302,20 @@ function applyOpenAICatalogPolicy(model: ApiModel<Api>, parsedModel: OpenAIModel
 	// Codex models: 400K figure includes output budget; input window is 272K.
 	if (parsedModel.variant.startsWith("codex") && parsedModel.variant !== "codex-spark") {
 		model.contextWindow = 272000;
+		return;
+	}
+	// GPT-5.4 mini/nano use plain OpenAI IDs on the Codex transport, but Codex still
+	// enforces the lower prompt budget for these variants. Codex discovery can also
+	// report inconsistent priorities for the GPT-5.4 family, so normalize by parsed
+	// variant instead of special-casing raw model ids.
+	if (model.api === "openai-codex-responses" && semverEqual(parsedModel.version, "5.4")) {
+		const normalizedPriority = CODEX_GPT_5_4_PRIORITY_BY_VARIANT[parsedModel.variant];
+		if (normalizedPriority !== undefined) {
+			model.priority = normalizedPriority;
+		}
+		if (parsedModel.variant === "mini" || parsedModel.variant === "nano") {
+			model.contextWindow = 272000;
+		}
 	}
 }
 
@@ -374,7 +394,10 @@ function inferAnthropicSupportedEfforts<TApi extends Api>(
 	parsedModel: AnthropicModel,
 	model: ApiModel<TApi>,
 ): readonly Effort[] {
-	if (model.api === "anthropic-messages" && semverGte(parsedModel.version, "4.6")) {
+	if (
+		(model.api === "anthropic-messages" || model.api === "bedrock-converse-stream") &&
+		semverGte(parsedModel.version, "4.6")
+	) {
 		return parsedModel.kind === "opus" ? DEFAULT_REASONING_EFFORTS_WITH_XHIGH : DEFAULT_REASONING_EFFORTS;
 	}
 	return inferFallbackEfforts(model);
@@ -427,6 +450,14 @@ function inferThinkingControlMode<TApi extends Api>(
 			return "budget";
 
 		case "bedrock-converse-stream":
+			if (parsedModel.family === "anthropic") {
+				if (semverGte(parsedModel.version, "4.6") && parsedModel.kind === "opus") {
+					return "anthropic-adaptive";
+				}
+				if (semverGte(parsedModel.version, "4.5")) {
+					return "anthropic-budget-effort";
+				}
+			}
 			return "budget";
 
 		default:
@@ -460,7 +491,7 @@ function parseGeminiModel(modelId: string): GeminiModel | null {
 }
 
 function parseAnthropicModel(modelId: string): AnthropicModel | null {
-	const match = /claude-(opus|sonnet)-(\d+(?:[.-]\d+){0,2})\b/.exec(modelId);
+	const match = /claude-(opus|sonnet)-(\d{1,2}(?:[.-]\d{1,2}){0,2})\b/.exec(modelId);
 	if (!match) {
 		return null;
 	}
@@ -472,7 +503,7 @@ 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|max|nano))?\b/.exec(modelId);
+	const match = /gpt-(\d+(?:\.\d+){0,2})(?:-(codex-spark|codex-mini|codex-max|codex|mini|max|nano))?\b/.exec(modelId);
 	if (!match) {
 		return null;
 	}