@oh-my-pi/pi-catalog 17.2.10 → 17.2.11

package/CHANGELOG.md

@@ -2,6 +2,14 @@
 
 ## [Unreleased]
 
+## [17.2.11] - 2026-08-07
+
+### Fixed
+
+- Increased the default stream idle-timeout floor on Amazon Bedrock to 900 seconds for reasoning and adaptive-thinking models (such as Claude) to prevent premature watchdog timeouts during long reasoning stretches.
+- Fixed Devin model families (including SWE-1.7, Claude 5, Gemini 3.6 Flash, Kimi K3, Grok 4.5, and Inkling) to correctly group as logical models with reasoning-effort routing instead of separate wire variants.
+- Added missing context-window and output-token limits for dynamically discovered Alibaba Token Plan models.
+
 ## [17.2.10] - 2026-08-06
 
 ### Changed

package/dist/types/compat/bedrock.d.ts

@@ -1,3 +1,3 @@
 import type { ModelSpec, ResolvedBedrockCompat } from "../types.js";
-/** Resolve Bedrock Converse prompt-cache capabilities once per model. */
+/** Resolve Bedrock Converse prompt-cache and stream-watchdog compat once per model. */
 export declare function buildBedrockCompat(spec: ModelSpec<"bedrock-converse-stream">): ResolvedBedrockCompat;

package/dist/types/provider-models/openai-compat.d.ts

@@ -342,6 +342,17 @@ export interface AlibabaCodingPlanModelManagerConfig {
 export declare function alibabaCodingPlanModelManagerOptions(config?: AlibabaCodingPlanModelManagerConfig): ModelManagerOptions<"openai-completions">;
 export { ALIBABA_TOKEN_PLAN_BASE_URL };
 export declare const ALIBABA_TOKEN_PLAN_STATIC_MODELS: readonly ModelSpec<"openai-completions">[];
+/**
+ * Metadata for Alibaba Token Plan models that are dynamically discovered but not
+ * in the static catalog. Context window and max tokens are sourced from
+ * official model documentation and provider catalogs. Unknown future models
+ * remain available with null limits instead of assigning unsafe guessed limits.
+ */
+export interface AlibabaTokenPlanModelLimits {
+    contextWindow: number;
+    maxTokens: number;
+}
+export declare const ALIBABA_TOKEN_PLAN_DISCOVERED_MODEL_LIMITS: Readonly<Record<string, AlibabaTokenPlanModelLimits>>;
 export interface AlibabaTokenPlanModelManagerConfig {
     apiKey?: string;
     baseUrl?: string;

package/dist/types/types.d.ts

@@ -478,6 +478,12 @@ export interface BedrockCompat {
      * Capability metadata only; zero means no explicit checkpoints.
      */
     promptCacheMaximumCheckpoints?: number;
+    /**
+     * Stream-watchdog idle-timeout fallback in ms; 0 disables the idle watchdog.
+     * Undefined defers to `PI_STREAM_IDLE_TIMEOUT_MS`, then the legacy
+     * `PI_OPENAI_STREAM_IDLE_TIMEOUT_MS` alias, then the 300s default.
+     */
+    streamIdleTimeoutMs?: number;
 }
 /** Fully-resolved Bedrock Converse prompt-cache capabilities, materialized once by `buildModel`. */
 export interface ResolvedBedrockCompat {
@@ -485,6 +491,13 @@ export interface ResolvedBedrockCompat {
     supportsLongPromptCacheRetention: boolean;
     promptCacheMinimumTokens: number;
     promptCacheMaximumCheckpoints: number;
+    /**
+     * Stream-watchdog idle-timeout fallback in ms for hosts with no keepalive
+     * events; 0 disables the idle watchdog. Undefined defers to
+     * `PI_STREAM_IDLE_TIMEOUT_MS`, then the legacy
+     * `PI_OPENAI_STREAM_IDLE_TIMEOUT_MS` alias, then the 300s default.
+     */
+    streamIdleTimeoutMs?: number;
 }
 /**
  * OpenRouter provider routing preferences.

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-catalog",
-	"version": "17.2.10",
+	"version": "17.2.11",
 	"description": "Model catalog for omp: bundled model database, provider discovery descriptors, model identity, classification, and equivalence",
 	"homepage": "https://omp.sh",
 	"author": "Can Boluk",
@@ -35,11 +35,11 @@
 	},
 	"dependencies": {
 		"@bufbuild/protobuf": "^2.12.1",
-		"@oh-my-pi/omptype": "17.2.10",
-		"@oh-my-pi/pi-utils": "17.2.10"
+		"@oh-my-pi/omptype": "17.2.11",
+		"@oh-my-pi/pi-utils": "17.2.11"
 	},
 	"devDependencies": {
-		"@oh-my-pi/pi-ai": "17.2.10",
+		"@oh-my-pi/pi-ai": "17.2.11",
 		"@types/bun": "^1.3.14"
 	},
 	"engines": {

package/src/compat/bedrock.ts

@@ -1,3 +1,4 @@
+import { supportsAdaptiveThinkingDisplay } from "../identity/family";
 import type { ModelSpec, ResolvedBedrockCompat } from "../types";
 import { applyCompatOverrides } from "./apply";
 
@@ -118,9 +119,35 @@ function detectedBedrockCompat(modelId: string): ResolvedBedrockCompat {
 	return NO_EXPLICIT_CHECKPOINTS;
 }
 
-/** Resolve Bedrock Converse prompt-cache capabilities once per model. */
+/**
+ * Bedrock ConverseStream sends no ping/keepalive events, so a reasoning model
+ * that goes quiet mid-thinking (summarized-display gaps, `omitted` thinking,
+ * or the wedged long tool-call generation of issue #4900) reads as a dead
+ * stream to the generic 300s idle watchdog and dies with "Provider stream
+ * stalled while waiting for the next event" (issue #4758's Bedrock variant).
+ * Widen the floor to 600s for reasoning models, mirroring the GLM coding-plan
+ * floor; explicit `spec.compat.streamIdleTimeoutMs` overrides still win.
+ */
+const BEDROCK_REASONING_STREAM_IDLE_TIMEOUT_MS = 600_000;
+/**
+ * Adaptive-thinking Claude (Opus 4.7+, Sonnet/Opus 5, Fable/Mythos 5) reasons
+ * for much longer stretches, and starting with Opus 4.7 / Fable 5 the
+ * Anthropic-side display default is `omitted` (issue #1373), so quiet gaps run
+ * longest on exactly this family — Fable 5 being the worst offender in the
+ * field. Direct Anthropic keeps these streams alive with ping keepalives and
+ * tolerates up to 3x the 300s idle budget of real-event silence (#4900);
+ * pingless Bedrock needs the same 900s tolerance in the raw idle floor.
+ */
+const BEDROCK_ADAPTIVE_THINKING_STREAM_IDLE_TIMEOUT_MS = 900_000;
+
+/** Resolve Bedrock Converse prompt-cache and stream-watchdog compat once per model. */
 export function buildBedrockCompat(spec: ModelSpec<"bedrock-converse-stream">): ResolvedBedrockCompat {
 	const compat = { ...detectedBedrockCompat(spec.id) };
+	compat.streamIdleTimeoutMs = spec.reasoning
+		? supportsAdaptiveThinkingDisplay(spec.id)
+			? BEDROCK_ADAPTIVE_THINKING_STREAM_IDLE_TIMEOUT_MS
+			: BEDROCK_REASONING_STREAM_IDLE_TIMEOUT_MS
+		: undefined;
 	applyCompatOverrides(compat, spec.compat);
 	return compat;
 }

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

@@ -2834,6 +2834,64 @@ export const ALIBABA_TOKEN_PLAN_STATIC_MODELS: readonly ModelSpec<"openai-comple
 	},
 ];
 
+/**
+ * Metadata for Alibaba Token Plan models that are dynamically discovered but not
+ * in the static catalog. Context window and max tokens are sourced from
+ * official model documentation and provider catalogs. Unknown future models
+ * remain available with null limits instead of assigning unsafe guessed limits.
+ */
+export interface AlibabaTokenPlanModelLimits {
+	contextWindow: number;
+	maxTokens: number;
+}
+
+export const ALIBABA_TOKEN_PLAN_DISCOVERED_MODEL_LIMITS: Readonly<Record<string, AlibabaTokenPlanModelLimits>> = {
+	"qwen3.6-plus": {
+		contextWindow: 1_000_000,
+		maxTokens: 65_536,
+	},
+	"qwen3.8-max": {
+		contextWindow: 1_000_000,
+		maxTokens: 131_072,
+	},
+	"deepseek-v4-flash": {
+		contextWindow: 1_000_000,
+		maxTokens: 384_000,
+	},
+	"deepseek-v4-flash-0731": {
+		contextWindow: 1_000_000,
+		maxTokens: 384_000,
+	},
+	"deepseek-v3.2": {
+		contextWindow: 131_072,
+		maxTokens: 65_536,
+	},
+	"glm-5.1": {
+		contextWindow: 202_752,
+		maxTokens: 128_000,
+	},
+	"glm-5": {
+		contextWindow: 202_752,
+		maxTokens: 16_384,
+	},
+	"kimi-k2.7-code": {
+		contextWindow: 262_144,
+		maxTokens: 262_144,
+	},
+	"kimi-k2.6": {
+		contextWindow: 262_144,
+		maxTokens: 262_144,
+	},
+	"kimi-k2.5": {
+		contextWindow: 262_144,
+		maxTokens: 98_304,
+	},
+	"minimax-m2.5": {
+		contextWindow: 196_608,
+		maxTokens: 32_768,
+	},
+};
+
 const ALIBABA_TOKEN_PLAN_NON_CHAT_MODEL_PREFIXES = [
 	"fun-asr",
 	"happyhorse-",
@@ -2888,10 +2946,19 @@ export function alibabaTokenPlanModelManagerOptions(
 								baseUrl: defaults.baseUrl,
 							};
 						}
-						// DeepSeek V4 family models discovered dynamically need reasoning config
-						if (defaults.id.startsWith("deepseek-v4")) {
+						const normalizedId = defaults.id.trim().toLowerCase();
+						const limits = ALIBABA_TOKEN_PLAN_DISCOVERED_MODEL_LIMITS[normalizedId];
+						const enriched = limits
+							? {
+									...defaults,
+									contextWindow: limits.contextWindow,
+									maxTokens: limits.maxTokens,
+								}
+							: defaults;
+
+						if (normalizedId.startsWith("deepseek-v4")) {
 							return {
-								...defaults,
+								...enriched,
 								reasoning: true,
 								thinking: {
 									mode: "effort" as const,
@@ -2899,7 +2966,7 @@ export function alibabaTokenPlanModelManagerOptions(
 								},
 							};
 						}
-						return defaults;
+						return enriched;
 					},
 					fetch: config?.fetch,
 				}),

package/src/types.ts

@@ -518,6 +518,12 @@ export interface BedrockCompat {
 	 * Capability metadata only; zero means no explicit checkpoints.
 	 */
 	promptCacheMaximumCheckpoints?: number;
+	/**
+	 * Stream-watchdog idle-timeout fallback in ms; 0 disables the idle watchdog.
+	 * Undefined defers to `PI_STREAM_IDLE_TIMEOUT_MS`, then the legacy
+	 * `PI_OPENAI_STREAM_IDLE_TIMEOUT_MS` alias, then the 300s default.
+	 */
+	streamIdleTimeoutMs?: number;
 }
 
 /** Fully-resolved Bedrock Converse prompt-cache capabilities, materialized once by `buildModel`. */
@@ -526,6 +532,13 @@ export interface ResolvedBedrockCompat {
 	supportsLongPromptCacheRetention: boolean;
 	promptCacheMinimumTokens: number;
 	promptCacheMaximumCheckpoints: number;
+	/**
+	 * Stream-watchdog idle-timeout fallback in ms for hosts with no keepalive
+	 * events; 0 disables the idle watchdog. Undefined defers to
+	 * `PI_STREAM_IDLE_TIMEOUT_MS`, then the legacy
+	 * `PI_OPENAI_STREAM_IDLE_TIMEOUT_MS` alias, then the 300s default.
+	 */
+	streamIdleTimeoutMs?: number;
 }
 
 /**

package/src/variant-collapse.ts

@@ -115,7 +115,7 @@ type DevinTierRoutes = Partial<Record<"off" | "minimal" | "low" | "medium" | "hi
 
 /** Devin families with a `-max` sibling: five wire tiers, `low` floor. */
 const DEVIN_FIVE_TIER_EFFORTS: readonly Effort[] = [Effort.Low, Effort.Medium, Effort.High, Effort.XHigh, Effort.Max];
-/** Devin families topping out at `-xhigh` (pre-5.6 GPT, 5.6 fast lanes). */
+/** Pre-5.6 Devin GPT families top out at `-xhigh`: four wire tiers, `low` floor. */
 const DEVIN_FOUR_TIER_EFFORTS: readonly Effort[] = [Effort.Low, Effort.Medium, Effort.High, Effort.XHigh];
 
 function devinTierFamily(
@@ -172,8 +172,7 @@ function devinTierFamily(
 
 /**
  * GPT-5.6 (Luna/Sol/Terra) serves per-tier siblings for the full five-tier
- * `low..max` wire scale; user efforts route 1:1 onto them. Devin serves no
- * `-max-priority` sibling, so the fast family tops out at `xhigh`.
+ * `low..max` wire scale in both standard and fast lanes.
  */
 function devinGpt56Families(variant: "luna" | "sol" | "terra", name: string): readonly EffortVariantFamily[] {
 	const base = `gpt-5-6-${variant}`;
@@ -200,8 +199,9 @@ function devinGpt56Families(variant: "luna" | "sol" | "terra", name: string): re
 				medium: `${base}-medium-priority`,
 				high: `${base}-high-priority`,
 				xhigh: `${base}-xhigh-priority`,
+				max: `${base}-max-priority`,
 			},
-			DEVIN_FOUR_TIER_EFFORTS,
+			DEVIN_FIVE_TIER_EFFORTS,
 		),
 	];
 }
@@ -378,6 +378,54 @@ export const GEMINI_CLI_VARIANT_COLLAPSE_TABLE: VariantCollapseTable = {
 };
 export const DEVIN_VARIANT_COLLAPSE_TABLE: VariantCollapseTable = {
 	families: [
+		devinTierFamily(
+			"claude-opus-5",
+			"Claude Opus 5",
+			{
+				low: "claude-opus-5-low",
+				medium: "claude-opus-5-medium",
+				high: "claude-opus-5-high",
+				xhigh: "claude-opus-5-xhigh",
+				max: "claude-opus-5-max",
+			},
+			DEVIN_FIVE_TIER_EFFORTS,
+		),
+		devinTierFamily(
+			"claude-opus-5-fast",
+			"Claude Opus 5 Fast",
+			{
+				low: "claude-opus-5-low-fast",
+				medium: "claude-opus-5-medium-fast",
+				high: "claude-opus-5-high-fast",
+				xhigh: "claude-opus-5-xhigh-fast",
+				max: "claude-opus-5-max-fast",
+			},
+			DEVIN_FIVE_TIER_EFFORTS,
+		),
+		devinTierFamily(
+			"claude-fable-5",
+			"Claude Fable 5",
+			{
+				low: "claude-5-fable-low",
+				medium: "claude-5-fable-medium",
+				high: "claude-5-fable-high",
+				xhigh: "claude-5-fable-xhigh",
+				max: "claude-5-fable-max",
+			},
+			DEVIN_FIVE_TIER_EFFORTS,
+		),
+		devinTierFamily(
+			"claude-sonnet-5",
+			"Claude Sonnet 5",
+			{
+				low: "claude-sonnet-5-low",
+				medium: "claude-sonnet-5-medium",
+				high: "claude-sonnet-5-high",
+				xhigh: "claude-sonnet-5-xhigh",
+				max: "claude-sonnet-5-max",
+			},
+			DEVIN_FIVE_TIER_EFFORTS,
+		),
 		devinTierFamily(
 			"claude-opus-4-7",
 			"Claude Opus 4.7",
@@ -522,6 +570,48 @@ export const DEVIN_VARIANT_COLLAPSE_TABLE: VariantCollapseTable = {
 		...devinGpt56Families("luna", "GPT-5.6 Luna"),
 		...devinGpt56Families("sol", "GPT-5.6 Sol"),
 		...devinGpt56Families("terra", "GPT-5.6 Terra"),
+		devinTierFamily(
+			"kimi-k3",
+			"Kimi K3",
+			{
+				low: "kimi-k3-low",
+				high: "kimi-k3-high",
+				max: "kimi-k3-max",
+			},
+			[Effort.Low, Effort.High, Effort.Max],
+		),
+		devinTierFamily(
+			"swe-1-7",
+			"SWE-1.7",
+			{
+				medium: "swe-1-7-medium",
+				max: "swe-1-7",
+			},
+			[Effort.Medium, Effort.Max],
+		),
+		devinTierFamily(
+			"grok-4-5",
+			"Grok 4.5",
+			{
+				low: "grok-4-5-low",
+				medium: "grok-4-5-medium",
+				high: "grok-4-5-high",
+			},
+			[Effort.Low, Effort.Medium, Effort.High],
+		),
+		devinTierFamily(
+			"inkling",
+			"Inkling",
+			{
+				off: "inkling-none",
+				low: "inkling-low",
+				medium: "inkling-medium",
+				high: "inkling-high",
+				xhigh: "inkling-xhigh",
+				max: "inkling-max",
+			},
+			DEVIN_FIVE_TIER_EFFORTS,
+		),
 		devinTierFamily(
 			"gemini-3-1-pro",
 			"Gemini 3.1 Pro",
@@ -542,6 +632,17 @@ export const DEVIN_VARIANT_COLLAPSE_TABLE: VariantCollapseTable = {
 			},
 			[Effort.Minimal, Effort.Low, Effort.Medium, Effort.High],
 		),
+		devinTierFamily(
+			"gemini-3-6-flash",
+			"Gemini 3.6 Flash",
+			{
+				minimal: "gemini-3-6-flash-minimal",
+				low: "gemini-3-6-flash-low",
+				medium: "gemini-3-6-flash-medium",
+				high: "gemini-3-6-flash-high",
+			},
+			[Effort.Minimal, Effort.Low, Effort.Medium, Effort.High],
+		),
 		devinTierFamily(
 			"gemini-3-flash",
 			"Gemini 3 Flash",