@kenkaiiii/gg-ai 5.90.0 → 5.91.1

package/dist/index.d.cts

@@ -241,6 +241,8 @@ interface StreamOptions {
     topP?: number;
     stop?: string[];
     thinking?: ThinkingLevel;
+    /** Effort rungs the endpoint itself declared for this model (synced gateway catalogs). */
+    thinkingLevels?: readonly ThinkingLevel[];
     apiKey?: string;
     baseUrl?: string;
     signal?: AbortSignal;

package/dist/index.d.ts

@@ -241,6 +241,8 @@ interface StreamOptions {
     topP?: number;
     stop?: string[];
     thinking?: ThinkingLevel;
+    /** Effort rungs the endpoint itself declared for this model (synced gateway catalogs). */
+    thinkingLevels?: readonly ThinkingLevel[];
     apiKey?: string;
     baseUrl?: string;
     signal?: AbortSignal;

package/dist/index.js

@@ -241,6 +241,8 @@ function formatError(err) {
         guidance: fastRejection.guidance
       };
     }
+    const miniMax = err.provider === "minimax" ? formatMiniMaxError(err, cleanMessage) : void 0;
+    if (miniMax) return miniMax;
     if (isUsageLimitError(err)) {
       const resetClause = err.resetsAt ? ` It resets at ${formatResetTime(err.resetsAt)}.` : "";
       return {
@@ -347,6 +349,38 @@ function inferSource(err) {
   }
   return "ggcoder";
 }
+function formatMiniMaxError(err, cleanMessage) {
+  const code = /\((\d{4,5})\)/.exec(err.message)?.[1];
+  const base = {
+    source: "provider",
+    message: cleanMessage,
+    provider: err.provider,
+    statusCode: err.statusCode,
+    ...err.requestId ? { requestId: err.requestId } : {}
+  };
+  if (code === "2049" || code === "1004") {
+    return {
+      ...base,
+      headline: "MiniMax didn't accept this API key.",
+      guidance: "A MiniMax key only works in the region that issued it (Global: platform.minimax.io, Mainland China: platform.minimaxi.com). Connect MiniMax again with the key so GG Coder can detect its region, and check the key is active."
+    };
+  }
+  if (code === "2061") {
+    return {
+      ...base,
+      headline: "Your MiniMax plan doesn't include this model.",
+      guidance: "Switch to MiniMax M3.1 Flash in the model selector \u2014 it's the model M Plan and Token Plan cover. MiniMax M3 needs a pay-as-you-go key."
+    };
+  }
+  if (code === "1008") {
+    return {
+      ...base,
+      headline: "MiniMax has no balance for this request.",
+      guidance: "Check this key's plan is active, or its pay-as-you-go balance, at platform.minimax.io (platform.minimaxi.com for Mainland China keys). Plan keys only cover MiniMax M3.1 Flash. Not a GG Coder issue."
+    };
+  }
+  return void 0;
+}
 function providerGuidance(provider, message, statusCode) {
   const name = provider ? providerDisplayName(provider) : "the provider";
   const status = provider ? PROVIDER_STATUS_URL[provider] : void 0;
@@ -1724,6 +1758,13 @@ function isAdaptiveThinkingModel(model) {
   return /opus-5|opus-4[-.]8|opus-4[-.]7|opus-4[-.]6|sonnet-5|haiku-5|fable-5|mythos-5/.test(model);
 }
 function toAnthropicThinking(level, maxTokens, model) {
+  if (model.startsWith("MiniMax-M3.1")) {
+    return {
+      thinking: { type: "adaptive" },
+      maxTokens,
+      outputConfig: { effort: level === "ultra" ? "max" : level }
+    };
+  }
   if (isAdaptiveThinkingModel(model)) {
     let effort = level;
     if (effort === "xhigh" && !/fable-5|mythos-5|opus-5|opus-4-8|opus-4-7|sonnet-5[-.]5|haiku-5[-.]5/.test(model)) {
@@ -1933,6 +1974,14 @@ function toCodexReasoningEffort(level, model) {
   if (level !== "ultra") return level;
   return model === "gpt-6-astra" || model === "gpt-6.1-sol" ? "xhigh" : "max";
 }
+var DECLARED_EFFORT_ORDER = ["low", "medium", "high", "xhigh", "max"];
+function effortForRequest(level, model, declared) {
+  const rungs = DECLARED_EFFORT_ORDER.filter((rung) => declared?.includes(rung));
+  if (rungs.length === 0) return toOpenAIReasoningEffort(level, model);
+  const wanted = level === "ultra" ? DECLARED_EFFORT_ORDER.length - 1 : DECLARED_EFFORT_ORDER.indexOf(level);
+  const atOrBelow = rungs.filter((rung) => DECLARED_EFFORT_ORDER.indexOf(rung) <= wanted);
+  return atOrBelow[atOrBelow.length - 1] ?? rungs[0];
+}
 function toOpenAIReasoningEffort(level, model) {
   const effort = level === "max" || level === "ultra" ? "xhigh" : level;
   if (model.startsWith("fugu") && (effort === "low" || effort === "medium")) {
@@ -2245,8 +2294,7 @@ async function* runStream(options) {
     ...system ? { system } : {},
     ...thinking ? { thinking } : {},
     ...outputConfig ? { output_config: outputConfig } : {},
-    ...options.temperature != null && !thinking ? { temperature: options.temperature } : {},
-    ...options.topP != null ? { top_p: options.topP } : {},
+    // API version 2026-10-21 (SDK 1.x) rejects temperature/top_p/top_k with a 400.
     ...options.stop ? { stop_sequences: options.stop } : {},
     ...(deferredPlan?.live ?? options.tools)?.length || options.serverTools?.length || options.webSearch ? (() => {
       const reservedServerNames = /* @__PURE__ */ new Set();
@@ -2347,7 +2395,7 @@ async function* runStream(options) {
       yield* synthesizeEventsFromMessage(message);
       return messageToResponse(message);
     } catch (err) {
-      throw toError(err, { fastMode: useFastMode });
+      throw toError(err, { fastMode: useFastMode, provider: options.provider });
     }
   }
   const contentParts = [];
@@ -2463,7 +2511,7 @@ async function* runStream(options) {
                 const rawPartial = accum.argsJson;
                 const snippet = rawPartial.length > 200 ? `${rawPartial.slice(0, 200)}\u2026` : rawPartial;
                 throw new ProviderError(
-                  "anthropic",
+                  options.provider,
                   `Tool "${accum.toolName}" input JSON was truncated in the stream (${rawPartial.length} bytes): ${snippet}; ${parseErr.message}`,
                   { cause: parseErr }
                 );
@@ -2566,15 +2614,15 @@ async function* runStream(options) {
       }
     }
   } catch (err) {
-    throw toError(err, { fastMode: useFastMode });
+    throw toError(err, { fastMode: useFastMode, provider: options.provider });
   }
   if (!receivedAnyEvent) {
-    throw new ProviderError("anthropic", "Stream ended without producing any events.", {
+    throw new ProviderError(options.provider, "Stream ended without producing any events.", {
       statusCode: 504
     });
   }
   if (stopReason === null) {
-    throw new ProviderError("anthropic", "Stream ended before completion (no stop_reason).", {
+    throw new ProviderError(options.provider, "Stream ended before completion (no stop_reason).", {
       statusCode: 504,
       cause: { partialContent: contentParts, outputTokens }
     });
@@ -2716,6 +2764,7 @@ function readUnifiedRateLimit(headers) {
   return { rejected: status === "rejected", ...resetsAt ? { resetsAt } : {} };
 }
 function toError(err, context) {
+  const provider = context?.provider ?? "anthropic";
   if (err instanceof ProviderError) return err;
   if (err instanceof Anthropic.APIError) {
     const errorBody = err.error;
@@ -2749,22 +2798,22 @@ function toError(err, context) {
     }
     if (isHardBillingMessage(message)) {
       const usageMessage = /usage limit reached/i.test(message) ? message : `usage limit reached: ${message}`;
-      return new ProviderError("anthropic", usageMessage, {
+      return new ProviderError(provider, usageMessage, {
         statusCode: err.status,
         ...requestId ? { requestId } : {},
         cause: err
       });
     }
-    return new ProviderError("anthropic", message, {
+    return new ProviderError(provider, message, {
       statusCode: err.status,
       ...requestId ? { requestId } : {},
       cause: err
     });
   }
   if (err instanceof Error) {
-    return new ProviderError("anthropic", err.message, { cause: err });
+    return new ProviderError(provider, err.message, { cause: err });
   }
-  return new ProviderError("anthropic", String(err));
+  return new ProviderError(provider, String(err));
 }
 
 // src/providers/openai.ts
@@ -2968,7 +3017,13 @@ async function* runStream2(options) {
     ...effectiveTemp != null && !options.thinking && !hasFixedKimiSampling ? { temperature: effectiveTemp } : {},
     ...options.topP != null && !hasFixedKimiSampling ? { top_p: options.topP } : {},
     ...options.stop ? { stop: options.stop } : {},
-    ...options.thinking && !usesThinkingParam && !isKimiK3 && !isKimiK27 && !isLocal ? { reasoning_effort: toOpenAIReasoningEffort(options.thinking, options.model) } : {},
+    ...options.thinking && !usesThinkingParam && !isKimiK3 && !isKimiK27 && !isLocal ? {
+      reasoning_effort: effortForRequest(
+        options.thinking,
+        options.model,
+        options.thinkingLevels
+      )
+    } : {},
     ...options.tools?.length ? {
       tools: toOpenAITools(options.tools, {
         strict: supportsStrictToolSampling(options.provider) && (options.strictTools ?? true)