@oh-my-pi/pi-catalog 16.1.16 → 16.1.17

package/CHANGELOG.md

@@ -2,6 +2,14 @@
 
 ## [Unreleased]
 
+## [16.1.17] - 2026-06-24
+
+### Fixed
+
+- Fixed the Umans GLM-5.2 thinking-level picker collapsing to a single `high` tier after dynamic discovery: the `max` upstream level now resolves to the internal `xhigh` effort, the picker shows both `high` and `xhigh`, and the metadata maps `xhigh` back to Umans's native `max` wire tier. ([#3192](https://github.com/can1357/oh-my-pi/issues/3192))
+- Fixed GitHub Copilot business and enterprise endpoints accepting image inputs that they reject with `400 vision is not supported`. The Copilot `/models` response advertises `capabilities.supports.vision = true` for Claude/GPT chat models on every host, but only the canonical personal endpoint (`https://api.githubcopilot.com`) actually serves them; `githubCopilotModelManagerOptions` now forces `input: ["text"]` whenever discovery resolves to a non-personal base URL, and `mergeDynamicModel` honours the dynamic value (instead of OR-upgrading) when the merged endpoint differs from the bundled reference. ([#3387](https://github.com/can1357/oh-my-pi/issues/3387))
+- Fixed OpenRouter Anthropic compat to strip Responses reasoning history during replay so signed thinking blocks are not sent back to routed Anthropic providers. ([#3399](https://github.com/can1357/oh-my-pi/issues/3399))
+
 ## [16.1.14] - 2026-06-22
 
 ### Added

package/dist/types/wire/github-copilot.d.ts

@@ -28,6 +28,16 @@ export type ParsedGitHubCopilotApiKey = {
     apiEndpoint?: string;
 };
 export declare function isPublicGitHubHost(host: string): boolean;
+/**
+ * Canonical personal-Copilot API host. The business
+ * (`api.business.githubcopilot.com`) and enterprise (`copilot-api.{domain}`)
+ * endpoints respond with HTTP 400 "vision is not supported" on image inputs,
+ * so catalog discovery and capability gates MUST honour the upstream's
+ * `supports.vision` flag only for this exact base URL.
+ */
+export declare const PERSONAL_GITHUB_COPILOT_BASE_URL: "https://api.githubcopilot.com";
+/** `true` when the resolved base URL is the canonical personal-Copilot host. */
+export declare function isPersonalGitHubCopilotBaseUrl(baseUrl: string | undefined): boolean;
 export declare function normalizeGitHubCopilotEnterpriseDomain(input: string | undefined): string | undefined;
 export declare function normalizeGitHubCopilotApiEndpoint(input: string | undefined): string | undefined;
 export declare function parseGitHubCopilotApiKey(apiKeyRaw: string): ParsedGitHubCopilotApiKey;

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-catalog",
-	"version": "16.1.16",
+	"version": "16.1.17",
 	"description": "Model catalog for omp: bundled model database, provider discovery descriptors, model identity, classification, and equivalence",
 	"homepage": "https://omp.sh",
 	"author": "Can Boluk",
@@ -34,12 +34,12 @@
 	},
 	"dependencies": {
 		"@bufbuild/protobuf": "^2.12.0",
-		"@oh-my-pi/pi-utils": "16.1.16",
+		"@oh-my-pi/pi-utils": "16.1.17",
 		"arktype": "^2.2.0",
 		"zod": "^4"
 	},
 	"devDependencies": {
-		"@oh-my-pi/pi-ai": "16.1.16",
+		"@oh-my-pi/pi-ai": "16.1.17",
 		"@types/bun": "^1.3.14"
 	},
 	"engines": {

package/src/compat/openai.ts

@@ -374,7 +374,7 @@ export function buildOpenAICompat(spec: ModelSpec<"openai-completions">): Resolv
 		reasoningDisableMode: resolveReasoningDisableMode(thinkingFormat),
 		omitReasoningEffort: false,
 		includeEncryptedReasoning: true,
-		filterReasoningHistory: false,
+		filterReasoningHistory: isOpenRouter && isAnthropicModel,
 		thinkingKeep: usesMoonshotKimiPreservedThinking ? "all" : undefined,
 		reasoningContentField: "reasoning_content",
 		// Backends that 400 follow-up requests when prior assistant tool-call turns lack `reasoning_content`:
@@ -480,6 +480,7 @@ export function buildOpenAIResponsesCompat(spec: OpenAIResponsesSpecLike): Resol
 	const id = spec.id ?? "";
 	const thinkingFormat: ResolvedOpenAISharedCompat["thinkingFormat"] = isOpenRouter ? "openrouter" : "openai";
 	const isKimiModel = id ? isKimiModelId(id) : false;
+	const isAnthropicModel = id ? isClaudeModelId(id) || isAnthropicNamespacedModelId(id) : false;
 	const isDeepseekFamily = id ? isDeepseekModelIdOrName(id) || isDeepseekModelIdOrName(spec.name) : false;
 	const reasoningCapable = Boolean(spec.reasoning);
 
@@ -505,7 +506,7 @@ export function buildOpenAIResponsesCompat(spec: OpenAIResponsesSpecLike): Resol
 		reasoningDisableMode: resolveReasoningDisableMode(thinkingFormat),
 		omitReasoningEffort: false,
 		includeEncryptedReasoning: spec.provider !== "xai-oauth",
-		filterReasoningHistory: spec.provider === "xai-oauth",
+		filterReasoningHistory: spec.provider === "xai-oauth" || (isOpenRouter && isAnthropicModel),
 		disableReasoningOnForcedToolChoice: isKimiModel,
 		disableReasoningOnToolChoice: isDeepseekFamily && reasoningCapable && !isOpenRouter,
 		supportsToolChoice: true,

package/src/model-manager.ts

@@ -349,7 +349,15 @@ function fingerprintStatic<TApi extends Api>(
 }
 
 function mergeDynamicModel<TApi extends Api>(existingModel: Model<TApi>, dynamicModel: Model<TApi>): Model<TApi> {
-	const supportsImage = existingModel.input.includes("image") || dynamicModel.input.includes("image");
+	// When discovery resolves the same model id to a different endpoint (e.g.
+	// a GitHub Copilot business/enterprise host), the bundled reference's
+	// capabilities are pinned to the canonical host and no longer apply —
+	// honour the dynamic value alone. Same-endpoint merges still OR-upgrade so
+	// a discovery that omits the capability flag doesn't drop bundled vision.
+	const endpointChanged = existingModel.baseUrl !== dynamicModel.baseUrl;
+	const supportsImage = endpointChanged
+		? dynamicModel.input.includes("image")
+		: existingModel.input.includes("image") || dynamicModel.input.includes("image");
 	// Re-build from spec stage: sparse compat comes from `compatConfig` (the
 	// verbatim override vocabulary), never the resolved `compat` record.
 	return buildModel({

package/src/model-thinking.ts

@@ -301,12 +301,12 @@ function getModelDefinedEfforts<TApi extends Api>(
 ): readonly Effort[] | undefined {
 	if (isGlm52ReasoningEffortModelId(spec.id)) {
 		// Z.ai/Zhipu and OpenRouter both surface GLM-5.2's full effort ladder,
-		// including the top `xhigh` (= "max") tier; Ollama Cloud exposes only
-		// high/xhigh.
+		// including the top `xhigh` (= "max") tier; Umans and Ollama Cloud
+		// expose only high/max.
 		if (isZaiThinkingFormat(compat) || isOpenRouterThinkingFormat(compat)) {
 			return DEFAULT_REASONING_EFFORTS_WITH_XHIGH;
 		}
-		if (isOllamaCloudGlm52ReasoningEffortModel(spec)) {
+		if (isUmansGlm52ReasoningEffortModel(spec) || isOllamaCloudGlm52ReasoningEffortModel(spec)) {
 			return GLM_52_HIGH_MAX_REASONING_EFFORTS;
 		}
 	}
@@ -325,6 +325,10 @@ function isOllamaCloudGlm52ReasoningEffortModel<TApi extends Api>(spec: ModelSpe
 	return spec.api === "ollama-chat" && spec.provider === "ollama-cloud" && isGlm52ReasoningEffortModelId(spec.id);
 }
 
+function isUmansGlm52ReasoningEffortModel<TApi extends Api>(spec: ModelSpec<TApi>): boolean {
+	return spec.api === "anthropic-messages" && spec.provider === "umans" && isGlm52ReasoningEffortModelId(spec.id);
+}
+
 function isMinimaxReasoningModelOnAnthropicEndpoint<TApi extends Api>(spec: ModelSpec<TApi>): boolean {
 	return spec.api === "anthropic-messages" && (isMinimaxM2FamilyModelId(spec.id) || isMinimaxM3FamilyModelId(spec.id));
 }
@@ -378,15 +382,16 @@ function inferDetectedEffortMap<TApi extends Api>(
 	//     `xhigh` 400s — collapse minimal->none, low/medium/high->high, xhigh->max.
 	//   - OpenRouter: `max` 400s and `xhigh` IS its max tier, so it passes `xhigh`
 	//     through literally (no map; the tier is exposed via getModelDefinedEfforts).
-	//   - Other openai-compat hosts (Fireworks, resellers) and Ollama Cloud keep
-	//     their distinct lower tiers and host quirks (e.g. Fireworks rejects
-	//     `minimal`, so `minimal->none` stays) and only remap the top `xhigh` UI
-	//     tier onto the genuine `max` budget. Filtered to supported efforts later.
+	//   - Umans and Ollama Cloud expose only high/max on their GLM-5.2 routes.
+	//   - Other openai-compat hosts (Fireworks, resellers) keep their distinct
+	//     lower tiers and host quirks (e.g. Fireworks rejects `minimal`, so
+	//     `minimal->none` stays) and only remap the top `xhigh` UI tier onto the
+	//     genuine `max` budget. Filtered to supported efforts later.
 	const isGlm52 = isGlm52ReasoningEffortModelId(spec.id);
 	if (isGlm52 && isZaiThinkingFormat(compat)) {
 		return ZAI_GLM_52_REASONING_EFFORT_MAP;
 	}
-	if (isOllamaCloudGlm52ReasoningEffortModel(spec)) {
+	if (isUmansGlm52ReasoningEffortModel(spec) || isOllamaCloudGlm52ReasoningEffortModel(spec)) {
 		return GLM_52_XHIGH_MAX_EFFORT_MAP;
 	}
 	if (isSakanaFuguReasoningModel(spec)) {
@@ -577,6 +582,9 @@ function inferThinkingControlMode<TApi extends Api>(
 			if (isMinimaxReasoningModelOnAnthropicEndpoint(spec)) {
 				return "anthropic-adaptive";
 			}
+			if (isUmansGlm52ReasoningEffortModel(spec)) {
+				return "anthropic-budget-effort";
+			}
 			if (parsedModel.family === "anthropic") {
 				if (semverGte(parsedModel.version, "4.6")) {
 					return "anthropic-adaptive";

package/src/models.json

@@ -71227,14 +71227,14 @@
 			"baseUrl": "https://api.code.umans.ai",
 			"reasoning": true,
 			"thinking": {
-				"mode": "budget",
+				"mode": "anthropic-budget-effort",
 				"efforts": [
-					"minimal",
-					"low",
-					"medium",
 					"high",
 					"xhigh"
-				]
+				],
+				"effortMap": {
+					"xhigh": "max"
+				}
 			},
 			"input": [
 				"text"

package/src/provider-models/openai-compat.ts

@@ -10,7 +10,12 @@ import type { ModelManagerOptions } from "../model-manager";
 import { getBundledModels } from "../models";
 import type { Api, FetchImpl, Model, ModelSpec, Provider, ThinkingConfig } from "../types";
 import { isAnthropicOAuthToken, isRecord, toBoolean, toNumber, toPositiveNumber } from "../utils";
-import { COPILOT_API_HEADERS, getGitHubCopilotBaseUrl, parseGitHubCopilotApiKey } from "../wire/github-copilot";
+import {
+	COPILOT_API_HEADERS,
+	getGitHubCopilotBaseUrl,
+	isPersonalGitHubCopilotBaseUrl,
+	parseGitHubCopilotApiKey,
+} from "../wire/github-copilot";
 import { createBundledReferenceMap, createReferenceResolver, toModelSpec } from "./bundled-references";
 
 const MODELS_DEV_URL = "https://models.dev/api.json";
@@ -566,7 +571,9 @@ const UMANS_REASONING_EFFORT_BY_LEVEL: Record<string, Effort> = {
 	medium: Effort.Medium,
 	high: Effort.High,
 	xhigh: Effort.XHigh,
+	max: Effort.XHigh,
 };
+const UMANS_MAX_REASONING_EFFORT_MAP = { [Effort.XHigh]: "max" } as const;
 const UMANS_DEFAULT_REASONING_EFFORTS = [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High, Effort.XHigh] as const;
 const UMANS_VIA_HANDOFF_MODEL_IDS = ["umans-glm-5.1", "umans-glm-5.2"] as const;
 
@@ -620,10 +627,20 @@ function mapUmansReasoningEfforts(value: unknown): readonly Effort[] {
 	return efforts.length > 0 ? efforts : UMANS_DEFAULT_REASONING_EFFORTS;
 }
 
+function umansHasMaxReasoningLevel(value: unknown): boolean {
+	return isRecord(value) && Array.isArray(value.levels) && value.levels.includes("max");
+}
+
 function mapUmansThinkingConfig(value: unknown): ThinkingConfig | undefined {
 	if (!umansReasoningSupported(value)) return undefined;
 	const efforts = mapUmansReasoningEfforts(value);
-	const thinking: ThinkingConfig = { mode: "budget", efforts };
+	const thinking: ThinkingConfig = {
+		mode: umansHasMaxReasoningLevel(value) ? "anthropic-budget-effort" : "budget",
+		efforts,
+	};
+	if (thinking.mode === "anthropic-budget-effort") {
+		thinking.effortMap = UMANS_MAX_REASONING_EFFORT_MAP;
+	}
 	if (isRecord(value)) {
 		if (value.can_disable === false) {
 			thinking.requiresEffort = true;
@@ -3113,10 +3130,16 @@ export function githubCopilotModelManagerOptions(config?: GithubCopilotModelMana
 								? entry.name
 								: (reference?.name ?? defaults.name);
 						const api = inferCopilotApi(defaults.id);
+						// `supports.vision` reports the model's intrinsic capability, but
+						// the business/enterprise endpoints respond `400 vision is not
+						// supported` on image inputs. Only honour the flag for the
+						// canonical personal-Copilot host.
 						const supportsVision = extractCopilotSupportsVision(entry);
-						const input: ModelSpec<Api>["input"] = supportsVision
-							? ["text", "image"]
-							: (reference?.input ?? defaults.input);
+						const input: ModelSpec<Api>["input"] = isPersonalGitHubCopilotBaseUrl(baseUrl)
+							? supportsVision
+								? ["text", "image"]
+								: (reference?.input ?? defaults.input)
+							: ["text"];
 						// With COPILOT_API_HEADERS the served window is the long-context
 						// ceiling; the default tier ends at token_prices.default.context_max
 						// prompt tokens. Cap the base entry to the default tier — the long

package/src/wire/github-copilot.ts

@@ -45,6 +45,20 @@ export function isPublicGitHubHost(host: string): boolean {
 	return PUBLIC_GITHUB_HOSTS.has(host.trim().toLowerCase());
 }
 
+/**
+ * Canonical personal-Copilot API host. The business
+ * (`api.business.githubcopilot.com`) and enterprise (`copilot-api.{domain}`)
+ * endpoints respond with HTTP 400 "vision is not supported" on image inputs,
+ * so catalog discovery and capability gates MUST honour the upstream's
+ * `supports.vision` flag only for this exact base URL.
+ */
+export const PERSONAL_GITHUB_COPILOT_BASE_URL = "https://api.githubcopilot.com" as const;
+
+/** `true` when the resolved base URL is the canonical personal-Copilot host. */
+export function isPersonalGitHubCopilotBaseUrl(baseUrl: string | undefined): boolean {
+	return baseUrl === PERSONAL_GITHUB_COPILOT_BASE_URL;
+}
+
 export function normalizeGitHubCopilotEnterpriseDomain(input: string | undefined): string | undefined {
 	const trimmed = input?.trim();
 	if (!trimmed) return undefined;