@oh-my-pi/pi-catalog 16.2.6 → 16.2.8

package/CHANGELOG.md

@@ -2,6 +2,13 @@
 
 ## [Unreleased]
 
+## [16.2.7] - 2026-06-30
+
+### Fixed
+
+- Fixed compatibility with Kimi K2.7 Code on native endpoints to ensure thinking mode is preserved and tool choice is not forced.
+- Fixed Cerebras gemma-4-31b dynamic discovery to correctly identify the model as image-capable, enabling proper serialization of attached images.
+
 ## [16.2.6] - 2026-06-29
 
 ### Fixed

package/dist/types/identity/family.d.ts

@@ -49,6 +49,17 @@ export declare const isMinimaxM3FamilyModelId: (modelId: string) => boolean;
 export declare const isOpenAIGptOssModelId: (modelId: string) => boolean;
 /** OpenAI model ids (gpt-*, o1-*, o3-*, o4-*, or prefixed with openai/). */
 export declare const isOpenAIModelId: (modelId: string) => boolean;
+/**
+ * OpenAI Codex models that honor `reasoning.context: "all_turns"` (full
+ * cross-turn reasoning replay). The `reasoning.context` field itself exists for
+ * the whole gpt-5/o-series family, but the `all_turns` value is only accepted
+ * from gpt-5.4 onward; earlier ids (`gpt-5.1-codex`, `gpt-5.3-codex`, and
+ * `gpt-5.3-codex-spark`) reject it with
+ * `Unsupported value: 'all_turns' is not supported with this model`. Version
+ * floor (not an allowlist) so 5.6/6.x inherit support automatically. Callers
+ * fall back to omitting `context`, letting the server default to `current_turn`.
+ */
+export declare const supportsAllTurnsReasoningContext: (modelId: string) => boolean;
 /**
  * Reasoning-capable GLM coding SKUs: glm-4.5 and up on the base / `-air` /
  * `-turbo` lines. Excludes the vision (`…v`) shape, the non-reasoning

package/dist/types/types.d.ts

@@ -381,6 +381,11 @@ export interface AnthropicCompat {
      * Default: auto-detected from provider/baseUrl and `model.reasoning`.
      */
     replayUnsignedThinking?: boolean;
+    /**
+     * Whether the endpoint requires `thinking.type: "enabled"` whenever the
+     * model reasons. Use for models that reject omitted or disabled thinking.
+     */
+    requiresThinkingEnabled?: boolean;
     /**
      * Prefix Anthropic built-in tool names (`web_search`, `code_execution`, ...)
      * when they are ordinary client tools. Some Anthropic-compatible gateways

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-catalog",
-	"version": "16.2.6",
+	"version": "16.2.8",
 	"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": "16.2.6",
+		"@oh-my-pi/pi-utils": "16.2.8",
 		"arktype": "^2.2.0",
 		"zod": "^4"
 	},
 	"devDependencies": {
-		"@oh-my-pi/pi-ai": "16.2.6",
+		"@oh-my-pi/pi-ai": "16.2.8",
 		"@types/bun": "^1.3.14"
 	},
 	"engines": {

package/src/compat/anthropic.ts

@@ -28,6 +28,14 @@ export function isOfficialAnthropicApiUrl(baseUrl?: string): boolean {
 	return lower === OFFICIAL_ANTHROPIC_URL || lower.startsWith(`${OFFICIAL_ANTHROPIC_URL}/`);
 }
 
+/** Mirrors `compat/openai.ts`; native-only host gating is the caller's responsibility. */
+const KIMI_K27_CODE_MODEL_PATTERN = /(?:^|\/)kimi[-._]?k2(?:[._-]?|p)7[-._]?code(?:[-._]?highspeed)?$/i;
+
+function matchesKimiK27CodeFamily(spec: ModelSpec<"anthropic-messages">): boolean {
+	if (KIMI_K27_CODE_MODEL_PATTERN.test(spec.id)) return true;
+	return spec.id === "kimi-for-coding" && /k2\.?7 code/i.test(spec.name ?? "");
+}
+
 /** Build the resolved anthropic-messages compat record for a model spec. */
 export function buildAnthropicCompat(spec: ModelSpec<"anthropic-messages">): ResolvedAnthropicCompat {
 	const baseUrl = spec.baseUrl;
@@ -40,6 +48,7 @@ export function buildAnthropicCompat(spec: ModelSpec<"anthropic-messages">): Res
 	// doesn't whitelist the `fine-grained-tool-streaming-2025-05-14` beta either
 	// (issue #2558), so eager tool-input streaming is unavailable on this host.
 	const isCopilot = modelMatchesHost(spec, "githubCopilot");
+	const requiresThinkingEnabled = modelMatchesHost(spec, "moonshotNative") && matchesKimiK27CodeFamily(spec);
 	const compat: ResolvedAnthropicCompat = {
 		officialEndpoint: official,
 		disableStrictTools: false,
@@ -53,12 +62,13 @@ export function buildAnthropicCompat(spec: ModelSpec<"anthropic-messages">): Res
 		// detection requires the canonical api.anthropic.com host plus a
 		// supported model id.
 		supportsMidConversationSystem: official && supportsMidConversationSystemMessages(spec.id),
-		supportsForcedToolChoice: !isAnthropicFableOrMythosModel(spec.id),
+		supportsForcedToolChoice: !requiresThinkingEnabled && !isAnthropicFableOrMythosModel(spec.id),
 		// Opus 4.7+ and Fable/Mythos reject temperature/top_p/top_k with a 400.
 		supportsSamplingParams: !hasOpus47ApiRestrictions(spec.id),
 		// Z.AI workaround (issue #814): its proxy deserializes tool_result blocks
 		// into a class that reads `.id`.
 		requiresToolResultId: isZai,
+		requiresThinkingEnabled,
 		// Official Anthropic enforces signature-based thinking-chain integrity, so
 		// unsigned thinking blocks must stay text there. Anthropic-compatible
 		// reasoning endpoints commonly emit unsigned thinking blocks while still

package/src/compat/openai.ts

@@ -39,6 +39,20 @@ const GLM_CODING_PLAN_STREAM_IDLE_TIMEOUT_MS = 600_000;
 const DEEPSEEK_REASONING_STREAM_IDLE_TIMEOUT_MS = 300_000;
 /** Kimi K2.6 can spend several minutes reasoning before the first visible token. */
 const KIMI_K26_REASONING_STREAM_IDLE_TIMEOUT_MS = 300_000;
+/**
+ * Native Kimi K2.7 Code requires `thinking.type: "enabled"` and rejects
+ * disabled thinking. Match the public id, its Fast variant, and the
+ * `kimi-code/kimi-for-coding` alias (which keeps the family name).
+ * Caller-disabled requests on non-native dialects (Fireworks `openai`,
+ * OpenRouter `openrouter`, …) MUST keep their per-dialect disable shape —
+ * gating on `isMoonshotKimi` is the caller's responsibility.
+ */
+const KIMI_K27_CODE_MODEL_PATTERN = /(?:^|\/)kimi[-._]?k2(?:[._-]?|p)7[-._]?code(?:[-._]?highspeed)?$/i;
+
+function matchesKimiK27CodeFamily(spec: ModelSpec<"openai-completions">): boolean {
+	if (KIMI_K27_CODE_MODEL_PATTERN.test(spec.id)) return true;
+	return spec.id === "kimi-for-coding" && /k2\.?7 code/i.test(spec.name ?? "");
+}
 /** Xiaomi MiMo Pro on api.xiaomimimo.com can stall ~2min before the first event (issue #1770). */
 const XIAOMI_MIMO_STREAM_IDLE_TIMEOUT_MS = 300_000;
 /** Alibaba Coding Plan (coding-intl.dashscope) qwen models idle before the first event (issue #1770). */
@@ -231,6 +245,7 @@ export function buildOpenAICompat(spec: ModelSpec<"openai-completions">): Resolv
 	const isKimiModel = isKimiModelId(spec.id);
 	const isMoonshotNative = modelMatchesHost(hostModel, "moonshotNative");
 	const isMoonshotKimi = isKimiModel && isMoonshotNative;
+	const requiresEnabledThinking = isMoonshotKimi && matchesKimiK27CodeFamily(spec);
 	const usesMoonshotKimiPreservedThinking = isMoonshotKimi && isKimiK26ModelId(spec.id);
 	const isAnthropicModel =
 		modelMatchesHost(hostModel, "anthropic") || isClaudeModelId(spec.id) || isAnthropicNamespacedModelId(spec.id);
@@ -403,7 +418,7 @@ export function buildOpenAICompat(spec: ModelSpec<"openai-completions">): Resolv
 		disableReasoningOnForcedToolChoice: isKimiModel || isAnthropicModel,
 		disableReasoningOnToolChoice: isDeepseekFamily && Boolean(spec.reasoning) && !isOpenRouter,
 		supportsToolChoice: !isDirectDeepseekReasoning,
-		supportsForcedToolChoice: true,
+		supportsForcedToolChoice: !requiresEnabledThinking,
 		supportsNamedToolChoice: provider !== "llama.cpp",
 		maxTokensField: useMaxTokens ? "max_tokens" : "max_completion_tokens",
 		requiresToolResultName: isMistral,
@@ -509,7 +524,9 @@ export function buildOpenAICompat(spec: ModelSpec<"openai-completions">): Resolv
 
 	applyCompatOverrides(compat, spec.compat);
 	if (spec.compat?.reasoningDisableMode === undefined) {
-		compat.reasoningDisableMode = resolveReasoningDisableMode(compat.thinkingFormat);
+		compat.reasoningDisableMode = requiresEnabledThinking
+			? "omit"
+			: resolveReasoningDisableMode(compat.thinkingFormat);
 	}
 	if (spec.compat?.omitReasoningEffort === undefined && !compat.supportsReasoningEffort) {
 		compat.omitReasoningEffort = true;

package/src/identity/family.ts

@@ -13,6 +13,7 @@ import {
 	parseAnthropicModel,
 	parseGlmModel,
 	parseKnownModel,
+	parseOpenAIModel,
 	semverGte,
 } from "./classify";
 
@@ -121,6 +122,22 @@ export const isOpenAIModelId = memo((modelId: string): boolean => {
 	return /(^|\/)(gpt|o1|o3|o4)[-.]/i.test(modelId) || modelId.toLowerCase().includes("openai/");
 });
 
+/**
+ * OpenAI Codex models that honor `reasoning.context: "all_turns"` (full
+ * cross-turn reasoning replay). The `reasoning.context` field itself exists for
+ * the whole gpt-5/o-series family, but the `all_turns` value is only accepted
+ * from gpt-5.4 onward; earlier ids (`gpt-5.1-codex`, `gpt-5.3-codex`, and
+ * `gpt-5.3-codex-spark`) reject it with
+ * `Unsupported value: 'all_turns' is not supported with this model`. Version
+ * floor (not an allowlist) so 5.6/6.x inherit support automatically. Callers
+ * fall back to omitting `context`, letting the server default to `current_turn`.
+ */
+export const supportsAllTurnsReasoningContext = memo((modelId: string): boolean => {
+	const parsed = parseOpenAIModel(bareModelId(modelId));
+	if (!parsed) return false;
+	return semverGte(parsed.version, "5.4");
+});
+
 /**
  * Reasoning-capable GLM coding SKUs: glm-4.5 and up on the base / `-air` /
  * `-turbo` lines. Excludes the vision (`…v`) shape, the non-reasoning

package/src/models.json

@@ -2476,7 +2476,7 @@
 			"api": "openai-completions",
 			"provider": "aimlapi",
 			"baseUrl": "https://api.aimlapi.com/v1",
-			"reasoning": true,
+			"reasoning": false,
 			"input": [
 				"text"
 			],
@@ -2487,24 +2487,7 @@
 				"cacheWrite": 0
 			},
 			"contextWindow": 163840,
-			"maxTokens": 16384,
-			"thinking": {
-				"mode": "effort",
-				"efforts": [
-					"minimal",
-					"low",
-					"medium",
-					"high",
-					"xhigh"
-				],
-				"effortMap": {
-					"minimal": "high",
-					"low": "high",
-					"medium": "high",
-					"high": "high",
-					"xhigh": "max"
-				}
-			}
+			"maxTokens": 16384
 		},
 		"deepseek/deepseek-chat-v3.1": {
 			"id": "deepseek/deepseek-chat-v3.1",
@@ -10620,6 +10603,35 @@
 				]
 			}
 		},
+		"xai.grok-4.3": {
+			"id": "xai.grok-4.3",
+			"name": "Grok 4.3",
+			"api": "bedrock-converse-stream",
+			"provider": "amazon-bedrock",
+			"baseUrl": "https://bedrock-runtime.us-east-1.amazonaws.com",
+			"reasoning": true,
+			"input": [
+				"text",
+				"image"
+			],
+			"cost": {
+				"input": 1.25,
+				"output": 2.5,
+				"cacheRead": 0.2,
+				"cacheWrite": 0
+			},
+			"contextWindow": 1000000,
+			"maxTokens": 131072,
+			"thinking": {
+				"mode": "budget",
+				"efforts": [
+					"minimal",
+					"low",
+					"medium",
+					"high"
+				]
+			}
+		},
 		"zai.glm-4.7": {
 			"id": "zai.glm-4.7",
 			"name": "GLM-4.7",
@@ -13672,7 +13684,7 @@
 			"api": "openai-completions",
 			"provider": "coreweave",
 			"baseUrl": "https://api.inference.wandb.ai/v1",
-			"reasoning": true,
+			"reasoning": false,
 			"input": [
 				"text"
 			],
@@ -13683,17 +13695,7 @@
 				"cacheWrite": 0
 			},
 			"contextWindow": 128000,
-			"maxTokens": 128000,
-			"thinking": {
-				"mode": "effort",
-				"efforts": [
-					"minimal",
-					"low",
-					"medium",
-					"high",
-					"xhigh"
-				]
-			}
+			"maxTokens": 128000
 		},
 		"meta-llama/Llama-3.3-70B-Instruct": {
 			"id": "meta-llama/Llama-3.3-70B-Instruct",
@@ -13701,7 +13703,7 @@
 			"api": "openai-completions",
 			"provider": "coreweave",
 			"baseUrl": "https://api.inference.wandb.ai/v1",
-			"reasoning": true,
+			"reasoning": false,
 			"input": [
 				"text"
 			],
@@ -13712,17 +13714,7 @@
 				"cacheWrite": 0
 			},
 			"contextWindow": 128000,
-			"maxTokens": 128000,
-			"thinking": {
-				"mode": "effort",
-				"efforts": [
-					"minimal",
-					"low",
-					"medium",
-					"high",
-					"xhigh"
-				]
-			}
+			"maxTokens": 128000
 		},
 		"meta-llama/Llama-4-Scout-17B-16E-Instruct": {
 			"id": "meta-llama/Llama-4-Scout-17B-16E-Instruct",
@@ -13730,7 +13722,7 @@
 			"api": "openai-completions",
 			"provider": "coreweave",
 			"baseUrl": "https://api.inference.wandb.ai/v1",
-			"reasoning": true,
+			"reasoning": false,
 			"input": [
 				"text",
 				"image"
@@ -13742,17 +13734,7 @@
 				"cacheWrite": 0
 			},
 			"contextWindow": 64000,
-			"maxTokens": 64000,
-			"thinking": {
-				"mode": "effort",
-				"efforts": [
-					"minimal",
-					"low",
-					"medium",
-					"high",
-					"xhigh"
-				]
-			}
+			"maxTokens": 64000
 		},
 		"microsoft/Phi-4-mini-instruct": {
 			"id": "microsoft/Phi-4-mini-instruct",
@@ -13760,7 +13742,7 @@
 			"api": "openai-completions",
 			"provider": "coreweave",
 			"baseUrl": "https://api.inference.wandb.ai/v1",
-			"reasoning": true,
+			"reasoning": false,
 			"input": [
 				"text"
 			],
@@ -13771,17 +13753,7 @@
 				"cacheWrite": 0
 			},
 			"contextWindow": 128000,
-			"maxTokens": 128000,
-			"thinking": {
-				"mode": "effort",
-				"efforts": [
-					"minimal",
-					"low",
-					"medium",
-					"high",
-					"xhigh"
-				]
-			}
+			"maxTokens": 128000
 		},
 		"MiniMaxAI/MiniMax-M2.5": {
 			"id": "MiniMaxAI/MiniMax-M2.5",
@@ -13789,7 +13761,7 @@
 			"api": "openai-completions",
 			"provider": "coreweave",
 			"baseUrl": "https://api.inference.wandb.ai/v1",
-			"reasoning": false,
+			"reasoning": true,
 			"input": [
 				"text"
 			],
@@ -13800,7 +13772,16 @@
 				"cacheWrite": 0
 			},
 			"contextWindow": 196608,
-			"maxTokens": 196608
+			"maxTokens": 196608,
+			"thinking": {
+				"mode": "effort",
+				"efforts": [
+					"low",
+					"medium",
+					"high"
+				],
+				"requiresEffort": true
+			}
 		},
 		"moonshotai/Kimi-K2.5": {
 			"id": "moonshotai/Kimi-K2.5",
@@ -13898,7 +13879,7 @@
 			"api": "openai-completions",
 			"provider": "coreweave",
 			"baseUrl": "https://api.inference.wandb.ai/v1",
-			"reasoning": false,
+			"reasoning": true,
 			"input": [
 				"text"
 			],
@@ -13909,7 +13890,17 @@
 				"cacheWrite": 0
 			},
 			"contextWindow": 262144,
-			"maxTokens": 262144
+			"maxTokens": 262144,
+			"thinking": {
+				"mode": "effort",
+				"efforts": [
+					"minimal",
+					"low",
+					"medium",
+					"high",
+					"xhigh"
+				]
+			}
 		},
 		"nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B": {
 			"id": "nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B",
@@ -21734,7 +21725,7 @@
 			"cost": {
 				"input": 0.075,
 				"output": 0.3,
-				"cacheRead": 0.037,
+				"cacheRead": 0,
 				"cacheWrite": 0
 			},
 			"contextWindow": 131072,
@@ -22420,13 +22411,13 @@
 				"text"
 			],
 			"cost": {
-				"input": 0,
-				"output": 0,
+				"input": 0.25,
+				"output": 0.69,
 				"cacheRead": 0,
 				"cacheWrite": 0
 			},
 			"contextWindow": 131072,
-			"maxTokens": 65536,
+			"maxTokens": 32768,
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
@@ -25048,7 +25039,7 @@
 			"api": "openai-completions",
 			"provider": "kilo",
 			"baseUrl": "https://api.kilo.ai/api/gateway",
-			"reasoning": true,
+			"reasoning": false,
 			"input": [
 				"text"
 			],
@@ -25059,24 +25050,7 @@
 				"cacheWrite": 0
 			},
 			"contextWindow": 163840,
-			"maxTokens": 65536,
-			"thinking": {
-				"mode": "effort",
-				"efforts": [
-					"minimal",
-					"low",
-					"medium",
-					"high",
-					"xhigh"
-				],
-				"effortMap": {
-					"minimal": "high",
-					"low": "high",
-					"medium": "high",
-					"high": "high",
-					"xhigh": "max"
-				}
-			}
+			"maxTokens": 65536
 		},
 		"deepseek/deepseek-chat-v3.1": {
 			"id": "deepseek/deepseek-chat-v3.1",
@@ -30816,6 +30790,7 @@
 				"supportsStrictMode": false,
 				"toolStrictMode": "mixed",
 				"stripDeepseekSpecialTokens": false,
+				"streamMarkupHealingPattern": "thinking",
 				"reasoningDeltasMayBeCumulative": false,
 				"emptyLengthFinishIsContextError": false,
 				"usesOpenAIToolCallIdLimit": false,
@@ -54444,6 +54419,36 @@
 				"requiresEffort": true
 			}
 		},
+		"minimaxai/minimax-m3": {
+			"id": "minimaxai/minimax-m3",
+			"name": "MiniMax-M3",
+			"api": "openai-completions",
+			"provider": "nvidia",
+			"baseUrl": "https://integrate.api.nvidia.com/v1",
+			"reasoning": true,
+			"input": [
+				"text",
+				"image"
+			],
+			"cost": {
+				"input": 0,
+				"output": 0,
+				"cacheRead": 0,
+				"cacheWrite": 0
+			},
+			"contextWindow": 1000000,
+			"maxTokens": 16384,
+			"thinking": {
+				"mode": "effort",
+				"efforts": [
+					"minimal",
+					"low",
+					"medium",
+					"high",
+					"xhigh"
+				]
+			}
+		},
 		"mistralai/codestral-22b-instruct-v0.1": {
 			"id": "mistralai/codestral-22b-instruct-v0.1",
 			"name": "Codestral 22b Instruct V0.1",
@@ -62195,9 +62200,9 @@
 				"image"
 			],
 			"cost": {
-				"input": 0.66,
-				"output": 3.41,
-				"cacheRead": 0.144,
+				"input": 0.55,
+				"output": 3.1999999999999997,
+				"cacheRead": 0.11,
 				"cacheWrite": 0
 			},
 			"contextWindow": 262144,
@@ -63522,7 +63527,7 @@
 			"api": "openrouter",
 			"baseUrl": "https://openrouter.ai/api/v1",
 			"provider": "openrouter",
-			"reasoning": true,
+			"reasoning": false,
 			"input": [
 				"text"
 			],
@@ -63533,22 +63538,7 @@
 				"cacheWrite": 0
 			},
 			"contextWindow": 163840,
-			"maxTokens": 16384,
-			"thinking": {
-				"mode": "effort",
-				"efforts": [
-					"minimal",
-					"low",
-					"medium",
-					"high"
-				],
-				"effortMap": {
-					"minimal": "high",
-					"low": "high",
-					"medium": "high",
-					"high": "high"
-				}
-			}
+			"maxTokens": 16384
 		},
 		"deepseek/deepseek-chat-v3.1": {
 			"id": "deepseek/deepseek-chat-v3.1",
@@ -65751,7 +65741,7 @@
 				"cacheWrite": 0
 			},
 			"contextWindow": 131072,
-			"maxTokens": 32768
+			"maxTokens": 100352
 		},
 		"moonshotai/kimi-k2-0905": {
 			"id": "moonshotai/kimi-k2-0905",
@@ -65770,7 +65760,7 @@
 				"cacheWrite": 0
 			},
 			"contextWindow": 262144,
-			"maxTokens": 262144
+			"maxTokens": 100352
 		},
 		"moonshotai/kimi-k2-0905:exacto": {
 			"id": "moonshotai/kimi-k2-0905:exacto",
@@ -65861,9 +65851,9 @@
 				"image"
 			],
 			"cost": {
-				"input": 0.66,
-				"output": 3.41,
-				"cacheRead": 0.144,
+				"input": 0.55,
+				"output": 3.1999999999999997,
+				"cacheRead": 0.11,
 				"cacheWrite": 0
 			},
 			"contextWindow": 262144,
@@ -69347,8 +69337,8 @@
 				"image"
 			],
 			"cost": {
-				"input": 0.28850000000000003,
-				"output": 2.65,
+				"input": 0.2596,
+				"output": 2.3850000000000002,
 				"cacheRead": 0.15,
 				"cacheWrite": 0
 			},
@@ -69758,13 +69748,13 @@
 				"text"
 			],
 			"cost": {
-				"input": 0.09,
+				"input": 0.09999999999999999,
 				"output": 0.3,
 				"cacheRead": 0.02,
 				"cacheWrite": 0
 			},
 			"contextWindow": 262144,
-			"maxTokens": 16384,
+			"maxTokens": 65536,
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
@@ -70846,8 +70836,8 @@
 				"text"
 			],
 			"cost": {
-				"input": 0.98,
-				"output": 3.08,
+				"input": 0.975,
+				"output": 4.300000000000001,
 				"cacheRead": 0.182,
 				"cacheWrite": 0
 			},
@@ -70874,7 +70864,7 @@
 				"text"
 			],
 			"cost": {
-				"input": 0.95,
+				"input": 0.94,
 				"output": 3,
 				"cacheRead": 0.18,
 				"cacheWrite": 0
@@ -71096,7 +71086,7 @@
 	"synthetic": {
 		"hf:MiniMaxAI/MiniMax-M3": {
 			"id": "hf:MiniMaxAI/MiniMax-M3",
-			"name": "MiniMaxAI/MiniMax-M3",
+			"name": "MiniMax-M3",
 			"api": "openai-completions",
 			"provider": "synthetic",
 			"baseUrl": "https://api.synthetic.new/openai/v1",
@@ -71111,7 +71101,7 @@
 				"cacheRead": 0.6,
 				"cacheWrite": 0
 			},
-			"contextWindow": 262144,
+			"contextWindow": 524288,
 			"maxTokens": 65536,
 			"thinking": {
 				"mode": "effort",
@@ -71126,7 +71116,7 @@
 		},
 		"hf:moonshotai/Kimi-K2.6": {
 			"id": "hf:moonshotai/Kimi-K2.6",
-			"name": "moonshotai/Kimi-K2.6",
+			"name": "Kimi K2.6",
 			"api": "openai-completions",
 			"provider": "synthetic",
 			"baseUrl": "https://api.synthetic.new/openai/v1",
@@ -71156,7 +71146,7 @@
 		},
 		"hf:nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4": {
 			"id": "hf:nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4",
-			"name": "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4",
+			"name": "Nemotron 3 Super 120B A12B",
 			"api": "openai-completions",
 			"provider": "synthetic",
 			"baseUrl": "https://api.synthetic.new/openai/v1",
@@ -71185,7 +71175,7 @@
 		},
 		"hf:openai/gpt-oss-120b": {
 			"id": "hf:openai/gpt-oss-120b",
-			"name": "openai/gpt-oss-120b",
+			"name": "GPT OSS 120B",
 			"api": "openai-completions",
 			"provider": "synthetic",
 			"baseUrl": "https://api.synthetic.new/openai/v1",
@@ -71196,7 +71186,7 @@
 			"cost": {
 				"input": 0.1,
 				"output": 0.1,
-				"cacheRead": 0,
+				"cacheRead": 0.1,
 				"cacheWrite": 0
 			},
 			"contextWindow": 131072,
@@ -71212,7 +71202,7 @@
 		},
 		"hf:Qwen/Qwen3.5-397B-A17B": {
 			"id": "hf:Qwen/Qwen3.5-397B-A17B",
-			"name": "Qwen/Qwen3.5-397B-A17B",
+			"name": "Qwen3.5 397B-A17B",
 			"api": "openai-completions",
 			"provider": "synthetic",
 			"baseUrl": "https://api.synthetic.new/openai/v1",
@@ -71223,7 +71213,7 @@
 			],
 			"cost": {
 				"input": 0.6,
-				"output": 3,
+				"output": 3.6,
 				"cacheRead": 0.6,
 				"cacheWrite": 0
 			},
@@ -71241,26 +71231,36 @@
 		},
 		"hf:Qwen/Qwen3.6-27B": {
 			"id": "hf:Qwen/Qwen3.6-27B",
-			"name": "Qwen/Qwen3.6-27B",
+			"name": "Qwen3.6 27B",
 			"api": "openai-completions",
 			"provider": "synthetic",
 			"baseUrl": "https://api.synthetic.new/openai/v1",
-			"reasoning": false,
+			"reasoning": true,
 			"input": [
-				"text"
+				"text",
+				"image"
 			],
 			"cost": {
-				"input": 0,
-				"output": 0,
-				"cacheRead": 0,
+				"input": 0.45,
+				"output": 3.6,
+				"cacheRead": 0.45,
 				"cacheWrite": 0
 			},
 			"contextWindow": 262144,
-			"maxTokens": 8192
+			"maxTokens": 65536,
+			"thinking": {
+				"mode": "effort",
+				"efforts": [
+					"minimal",
+					"low",
+					"medium",
+					"high"
+				]
+			}
 		},
 		"hf:zai-org/GLM-4.7": {
 			"id": "hf:zai-org/GLM-4.7",
-			"name": "zai-org/GLM-4.7",
+			"name": "GLM-4.7",
 			"api": "openai-completions",
 			"provider": "synthetic",
 			"baseUrl": "https://api.synthetic.new/openai/v1",
@@ -71269,13 +71269,13 @@
 				"text"
 			],
 			"cost": {
-				"input": 0.55,
+				"input": 0.45,
 				"output": 2.19,
-				"cacheRead": 0,
+				"cacheRead": 0.45,
 				"cacheWrite": 0
 			},
 			"contextWindow": 202752,
-			"maxTokens": 64000,
+			"maxTokens": 65536,
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
@@ -71289,7 +71289,7 @@
 		},
 		"hf:zai-org/GLM-4.7-Flash": {
 			"id": "hf:zai-org/GLM-4.7-Flash",
-			"name": "zai-org/GLM-4.7-Flash",
+			"name": "GLM-4.7-Flash",
 			"api": "openai-completions",
 			"provider": "synthetic",
 			"baseUrl": "https://api.synthetic.new/openai/v1",
@@ -71298,9 +71298,9 @@
 				"text"
 			],
 			"cost": {
-				"input": 0.06,
-				"output": 0.4,
-				"cacheRead": 0.06,
+				"input": 0.1,
+				"output": 0.5,
+				"cacheRead": 0.1,
 				"cacheWrite": 0
 			},
 			"contextWindow": 196608,
@@ -71318,7 +71318,7 @@
 		},
 		"hf:zai-org/GLM-5.1": {
 			"id": "hf:zai-org/GLM-5.1",
-			"name": "zai-org/GLM-5.1",
+			"name": "GLM-5.1",
 			"api": "openai-completions",
 			"provider": "synthetic",
 			"baseUrl": "https://api.synthetic.new/openai/v1",
@@ -71347,22 +71347,32 @@
 		},
 		"hf:zai-org/GLM-5.2": {
 			"id": "hf:zai-org/GLM-5.2",
-			"name": "zai-org/GLM-5.2",
+			"name": "GLM-5.2",
 			"api": "openai-completions",
 			"provider": "synthetic",
 			"baseUrl": "https://api.synthetic.new/openai/v1",
-			"reasoning": false,
+			"reasoning": true,
 			"input": [
 				"text"
 			],
 			"cost": {
-				"input": 0,
-				"output": 0,
-				"cacheRead": 0,
+				"input": 1.4,
+				"output": 4.4,
+				"cacheRead": 1.4,
 				"cacheWrite": 0
 			},
 			"contextWindow": 524288,
-			"maxTokens": 8192
+			"maxTokens": 65536,
+			"thinking": {
+				"mode": "effort",
+				"efforts": [
+					"minimal",
+					"low",
+					"medium",
+					"high",
+					"xhigh"
+				]
+			}
 		},
 		"syn:large:text": {
 			"id": "syn:large:text",
@@ -71682,7 +71692,7 @@
 			"api": "openai-completions",
 			"provider": "together",
 			"baseUrl": "https://api.together.xyz/v1",
-			"reasoning": true,
+			"reasoning": false,
 			"input": [
 				"text",
 				"image"
@@ -71694,17 +71704,7 @@
 				"cacheWrite": 0.18
 			},
 			"contextWindow": 10000000,
-			"maxTokens": 32768,
-			"thinking": {
-				"mode": "effort",
-				"efforts": [
-					"minimal",
-					"low",
-					"medium",
-					"high",
-					"xhigh"
-				]
-			}
+			"maxTokens": 32768
 		},
 		"MiniMaxAI/MiniMax-M2.5": {
 			"id": "MiniMaxAI/MiniMax-M2.5",
@@ -72381,6 +72381,38 @@
 				"escapeBuiltinToolNames": true
 			}
 		},
+		"umans-glm-5.2-nvfp4": {
+			"id": "umans-glm-5.2-nvfp4",
+			"name": "Umans GLM 5.2 NVFP4 (experimental, short test from Jun 29)",
+			"api": "anthropic-messages",
+			"provider": "umans",
+			"baseUrl": "https://api.code.umans.ai",
+			"reasoning": true,
+			"thinking": {
+				"mode": "anthropic-budget-effort",
+				"efforts": [
+					"high",
+					"xhigh"
+				],
+				"effortMap": {
+					"xhigh": "max"
+				}
+			},
+			"input": [
+				"text"
+			],
+			"cost": {
+				"input": 0,
+				"output": 0,
+				"cacheRead": 0,
+				"cacheWrite": 0
+			},
+			"contextWindow": 405504,
+			"maxTokens": 131071,
+			"compat": {
+				"escapeBuiltinToolNames": true
+			}
+		},
 		"umans-kimi-k2.7": {
 			"id": "umans-kimi-k2.7",
 			"name": "Umans Kimi K2.7 Code",
@@ -75747,6 +75779,7 @@
 				"supportsForcedToolChoice": true,
 				"supportsSamplingParams": true,
 				"requiresToolResultId": false,
+				"requiresThinkingEnabled": false,
 				"replayUnsignedThinking": false,
 				"escapeBuiltinToolNames": false
 			}
@@ -80633,7 +80666,7 @@
 		},
 		"zai/glm-4.5": {
 			"id": "zai/glm-4.5",
-			"name": "GLM-4.5",
+			"name": "GLM 4.5",
 			"api": "anthropic-messages",
 			"baseUrl": "https://ai-gateway.vercel.sh",
 			"provider": "vercel-ai-gateway",
@@ -82062,6 +82095,14 @@
 			},
 			"contextWindow": 2000000,
 			"maxTokens": 2000000,
+			"compat": {
+				"reasoningEffortMap": {
+					"minimal": "low"
+				},
+				"includeEncryptedReasoning": false,
+				"filterReasoningHistory": true,
+				"omitReasoningEffort": false
+			},
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
@@ -82074,14 +82115,6 @@
 				"effortMap": {
 					"minimal": "low"
 				}
-			},
-			"compat": {
-				"reasoningEffortMap": {
-					"minimal": "low"
-				},
-				"includeEncryptedReasoning": false,
-				"filterReasoningHistory": true,
-				"omitReasoningEffort": false
 			}
 		},
 		"grok-4.3": {
@@ -82103,6 +82136,14 @@
 			},
 			"contextWindow": 1000000,
 			"maxTokens": 1000000,
+			"compat": {
+				"reasoningEffortMap": {
+					"minimal": "low"
+				},
+				"includeEncryptedReasoning": false,
+				"filterReasoningHistory": true,
+				"omitReasoningEffort": false
+			},
 			"thinking": {
 				"mode": "effort",
 				"efforts": [
@@ -82115,14 +82156,6 @@
 				"effortMap": {
 					"minimal": "low"
 				}
-			},
-			"compat": {
-				"reasoningEffortMap": {
-					"minimal": "low"
-				},
-				"includeEncryptedReasoning": false,
-				"filterReasoningHistory": true,
-				"omitReasoningEffort": false
 			}
 		},
 		"grok-build": {
@@ -82174,12 +82207,12 @@
 			"contextWindow": 256000,
 			"maxTokens": 256000,
 			"compat": {
-				"includeEncryptedReasoning": false,
-				"filterReasoningHistory": true,
-				"omitReasoningEffort": true,
 				"reasoningEffortMap": {
 					"minimal": "low"
 				},
+				"includeEncryptedReasoning": false,
+				"filterReasoningHistory": true,
+				"omitReasoningEffort": true,
 				"supportsReasoningEffort": false
 			}
 		},
@@ -82223,9 +82256,9 @@
 				"text"
 			],
 			"cost": {
-				"input": 0.1,
-				"output": 0.3,
-				"cacheRead": 0.01,
+				"input": 0.14,
+				"output": 0.28,
+				"cacheRead": 0.0028,
 				"cacheWrite": 0
 			},
 			"contextWindow": 262144,
@@ -82259,9 +82292,9 @@
 				"image"
 			],
 			"cost": {
-				"input": 0.4,
-				"output": 2,
-				"cacheRead": 0.08,
+				"input": 0.14,
+				"output": 0.28,
+				"cacheRead": 0.0028,
 				"cacheWrite": 0
 			},
 			"contextWindow": 262144,
@@ -82294,9 +82327,9 @@
 				"text"
 			],
 			"cost": {
-				"input": 1,
-				"output": 3,
-				"cacheRead": 0.2,
+				"input": 0.435,
+				"output": 0.87,
+				"cacheRead": 0.0036,
 				"cacheWrite": 0
 			},
 			"contextWindow": 1048576,
@@ -82330,9 +82363,9 @@
 				"image"
 			],
 			"cost": {
-				"input": 0.4,
-				"output": 2,
-				"cacheRead": 0.08,
+				"input": 0.14,
+				"output": 0.28,
+				"cacheRead": 0.0028,
 				"cacheWrite": 0
 			},
 			"contextWindow": 1048576,
@@ -82365,9 +82398,9 @@
 				"text"
 			],
 			"cost": {
-				"input": 1,
-				"output": 3,
-				"cacheRead": 0.2,
+				"input": 0.435,
+				"output": 0.87,
+				"cacheRead": 0.0036,
 				"cacheWrite": 0
 			},
 			"contextWindow": 1048576,

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

@@ -803,6 +803,18 @@ export function groqModelManagerOptions(config?: GroqModelManagerConfig): ModelM
 // 3. Cerebras
 // ---------------------------------------------------------------------------
 
+const CEREBRAS_IMAGE_INPUT_MODEL_IDS = new Set(["gemma-4-31b"]);
+
+function applyCerebrasDiscoveryOverrides(model: ModelSpec<"openai-completions">): ModelSpec<"openai-completions"> {
+	if (!CEREBRAS_IMAGE_INPUT_MODEL_IDS.has(model.id)) {
+		return model;
+	}
+	return {
+		...model,
+		input: ["text", "image"],
+	};
+}
+
 export interface CerebrasModelManagerConfig {
 	apiKey?: string;
 	baseUrl?: string;
@@ -812,7 +824,27 @@ export interface CerebrasModelManagerConfig {
 export function cerebrasModelManagerOptions(
 	config?: CerebrasModelManagerConfig,
 ): ModelManagerOptions<"openai-completions"> {
-	return createSimpleOpenAICompletionsOptions("cerebras", "https://api.cerebras.ai/v1", config);
+	const apiKey = config?.apiKey;
+	const baseUrl = config?.baseUrl ?? "https://api.cerebras.ai/v1";
+	const references = createBundledReferenceMap<"openai-completions">("cerebras");
+	return {
+		providerId: "cerebras",
+		...(apiKey && {
+			fetchDynamicModels: () =>
+				fetchOpenAICompatibleModels({
+					api: "openai-completions",
+					provider: "cerebras",
+					baseUrl,
+					apiKey,
+					mapModel: (entry, defaults) => {
+						const reference = references.get(defaults.id);
+						const model = mapWithBundledReference(entry, defaults, reference);
+						return applyCerebrasDiscoveryOverrides(model);
+					},
+					fetch: config?.fetch,
+				}),
+		}),
+	};
 }
 
 // ---------------------------------------------------------------------------
@@ -1294,7 +1326,7 @@ export function clampFireworksKimiMaxTokens(modelId: string, candidate: number |
 export const KIMI_K27_CODE_RECOMMENDED_MAX_TOKENS = 32_768;
 
 export function isKimiK27CodeModelId(modelId: string): boolean {
-	return /(?:^|\/)kimi[-._]?k2(?:[._-]?|p)7[-._]?code$/i.test(modelId);
+	return /(?:^|\/)kimi[-._]?k2(?:[._-]?|p)7[-._]?code(?:[-._]?highspeed)?$/i.test(modelId);
 }
 
 export function clampKimiK27CodeMaxTokens(modelId: string, candidate: number): number;

package/src/types.ts

@@ -420,6 +420,11 @@ export interface AnthropicCompat {
 	 * Default: auto-detected from provider/baseUrl and `model.reasoning`.
 	 */
 	replayUnsignedThinking?: boolean;
+	/**
+	 * Whether the endpoint requires `thinking.type: "enabled"` whenever the
+	 * model reasons. Use for models that reject omitted or disabled thinking.
+	 */
+	requiresThinkingEnabled?: boolean;
 	/**
 	 * Prefix Anthropic built-in tool names (`web_search`, `code_execution`, ...)
 	 * when they are ordinary client tools. Some Anthropic-compatible gateways