EyJunge1 · Oct 21, 2026 · Oct 20, 2026 · Oct 21, 2026
diff --git a/README.md b/README.md
@@ -482,7 +482,7 @@ Manual `memoryProvider` modes:
 - `openai-chat`: OpenAI Chat Completions compatible API with tool/function calling. This can work with compatible proxies such as LiteLLM only when the selected upstream model and proxy preserve tool calls.
 - `openai-responses`: OpenAI Responses API with function-call output.
 - `anthropic`: Anthropic Messages API with tool use.
-- `minimax`: MiniMax Anthropic Messages-compatible endpoint. Set `memoryApiUrl` to the global endpoint (`https://api.minimax.io`) or the China endpoint (`https://api.minimaxi.com`); the `/anthropic/v1/messages` path and `x-api-key` header are applied automatically. Current models include `MiniMax-M3` (1,000,000-token context; adaptive or disabled thinking) and `MiniMax-M2.7` (204,800-token context; always-on thinking). `MiniMax-M3` supports adaptive thinking through `memoryExtraParams`.
+- `minimax`: MiniMax Anthropic Messages-compatible endpoint. Set `memoryApiUrl` to the global endpoint (`https://api.minimax.io`) or the [China endpoint](https://platform.minimaxi.com/docs/api-reference/text-anthropic-api.md) (`https://api.minimax.cn`); the `/anthropic/v1/messages` path and `x-api-key` header are applied automatically. Use `MiniMax-M3.1-Flash` (recommended) or `MiniMax-M3`, both with a 1,000,000-token context. Flash always thinks; optionally set `memoryExtraParams` to `{ "output_config": { "effort": "max" } }` with `low`, `medium`, `high`, `xhigh`, or `max` (default). Flash rejects disabled thinking and effort `none`, and does not support `service_tier`. M3 thinking is off by default on this endpoint; enable it with `{ "thinking": { "type": "adaptive" } }`. Model IDs remain explicitly configured; existing custom or legacy IDs are passed through unchanged. See the [MiniMax compatibility reference](https://platform.minimax.io/docs/api-reference/text-anthropic-api.md).
 - `orcarouter`: OpenAI-compatible model gateway with namespaced model IDs. `memoryApiUrl` and `memoryModel` are optional — they default to `https://api.orcarouter.ai/v1` and `orcarouter/auto` (a routing alias that selects a capable model per request). If you set `memoryModel`, use a namespaced ID such as `openai/gpt-5.5` or `deepseek/deepseek-v4-flash`; OrcaRouter rejects bare model names. Example:
   ```jsonc
   "memoryProvider": "orcarouter",

diff --git a/i18n/README.ar.md b/i18n/README.ar.md
@@ -469,7 +469,7 @@ memory({ mode: "import", inputPath: "./memories.json" });
 - `openai-chat`: API متوافق مع OpenAI Chat Completions مع استدعاء أدوات/دوال. يمكن أن يعمل مع وكلاء متوافقين مثل LiteLLM فقط عندما يحافظ النموذج العلوي والوكيل المحددان على استدعاءات الأدوات.
 - `openai-responses`: OpenAI Responses API مع مخرجات استدعاء دوال.
 - `anthropic`: Anthropic Messages API مع استخدام الأدوات.
-- `minimax`: نقطة نهاية MiniMax متوافقة مع Anthropic Messages. عيّن `memoryApiUrl` إلى النقطة العالمية (`https://api.minimax.io`) أو نقطة الصين (`https://api.minimaxi.com`)؛ يُطبَّق مسار `/anthropic/v1/messages` وترويسة `x-api-key` تلقائيًا. النماذج الحالية تشمل `MiniMax-M3` (سياق 1,000,000 رمز؛ تفكير تكيفي أو معطّل) و`MiniMax-M2.7` (سياق 204,800 رمز؛ تفكير دائم التشغيل). يدعم `MiniMax-M3` التفكير التكيفي عبر `memoryExtraParams`.
+- `minimax`: نقطة نهاية MiniMax متوافقة مع Anthropic Messages. عيّن `memoryApiUrl` إلى النقطة العالمية (`https://api.minimax.io`) أو [نقطة الصين](https://platform.minimaxi.com/docs/api-reference/text-anthropic-api.md) (`https://api.minimax.cn`)؛ يُطبَّق مسار `/anthropic/v1/messages` وترويسة `x-api-key` تلقائيًا. استخدم `MiniMax-M3.1-Flash` (موصى به) أو `MiniMax-M3`، وكلاهما بسياق 1,000,000 رمز. Flash يفكر دائمًا؛ ويمكن اختياريًا تعيين `memoryExtraParams` إلى `{ "output_config": { "effort": "max" } }` بالقيم `low` أو `medium` أو `high` أو `xhigh` أو `max` (الافتراضي). يرفض Flash تعطيل التفكير وeffort `none`، ولا يدعم `service_tier`. تفكير M3 معطّل افتراضيًا على هذه النقطة؛ فعّله بـ `{ "thinking": { "type": "adaptive" } }`. تبقى معرّفات النماذج مُعدّة صراحةً؛ وتُمرَّر المعرّفات المخصصة أو القديمة دون تغيير. راجع [مرجع توافق MiniMax](https://platform.minimax.io/docs/api-reference/text-anthropic-api.md).
 - `orcarouter`: بوابة نماذج متوافقة مع OpenAI بمعرّفات نماذج ذات مساحة أسماء. `memoryApiUrl` و`memoryModel` اختياريان — الافتراضي `https://api.orcarouter.ai/v1` و`orcarouter/auto` (اسم مستعار للتوجيه يختار نموذجًا قادرًا لكل طلب). إذا عيّنت `memoryModel`، استخدم معرّفًا ذا مساحة أسماء مثل `openai/gpt-5.5` أو `deepseek/deepseek-v4-flash`؛ يرفض OrcaRouter أسماء النماذج العارية. مثال:
   ```jsonc
   "memoryProvider": "orcarouter",

diff --git a/i18n/README.de.md b/i18n/README.de.md
@@ -469,7 +469,7 @@ Manuelle `memoryProvider`-Modi:
 - `openai-chat`: OpenAI-Chat-Completions-kompatible API mit Tool-/Function-Calling. Kann mit kompatiblen Proxies wie LiteLLM nur funktionieren, wenn das gewählte Upstream-Modell und der Proxy Tool-Calls erhalten.
 - `openai-responses`: OpenAI Responses API mit Function-Call-Ausgabe.
 - `anthropic`: Anthropic Messages API mit Tool Use.
-- `minimax`: MiniMax Anthropic-Messages-kompatibler Endpoint. `memoryApiUrl` auf den globalen Endpoint (`https://api.minimax.io`) oder den China-Endpoint (`https://api.minimaxi.com`) setzen; der Pfad `/anthropic/v1/messages` und der Header `x-api-key` werden automatisch angewendet. Aktuelle Modelle umfassen `MiniMax-M3` (1.000.000-Token-Kontext; adaptives oder deaktiviertes Thinking) und `MiniMax-M2.7` (204.800-Token-Kontext; Thinking immer an). `MiniMax-M3` unterstützt adaptives Thinking über `memoryExtraParams`.
+- `minimax`: MiniMax Anthropic-Messages-kompatibler Endpoint. `memoryApiUrl` auf den globalen Endpoint (`https://api.minimax.io`) oder den [China-Endpoint](https://platform.minimaxi.com/docs/api-reference/text-anthropic-api.md) (`https://api.minimax.cn`) setzen; der Pfad `/anthropic/v1/messages` und der Header `x-api-key` werden automatisch angewendet. `MiniMax-M3.1-Flash` (empfohlen) oder `MiniMax-M3` verwenden, beide mit 1.000.000-Token-Kontext. Flash denkt immer; optional `memoryExtraParams` auf `{ "output_config": { "effort": "max" } }` setzen mit `low`, `medium`, `high`, `xhigh` oder `max` (Standard). Flash lehnt deaktiviertes Thinking und Effort `none` ab und unterstützt kein `service_tier`. M3-Thinking ist auf diesem Endpoint standardmäßig aus; aktivieren mit `{ "thinking": { "type": "adaptive" } }`. Modell-IDs bleiben explizit konfiguriert; vorhandene Custom- oder Legacy-IDs werden unverändert durchgereicht. Siehe die [MiniMax-Kompatibilitätsreferenz](https://platform.minimax.io/docs/api-reference/text-anthropic-api.md).
 - `orcarouter`: OpenAI-kompatibles Model-Gateway mit namenspaced Model-IDs. `memoryApiUrl` und `memoryModel` sind optional — sie defaulten auf `https://api.orcarouter.ai/v1` und `orcarouter/auto` (ein Routing-Alias, der pro Request ein fähiges Modell wählt). Wenn du `memoryModel` setzt, eine namenspaced ID wie `openai/gpt-5.5` oder `deepseek/deepseek-v4-flash` verwenden; OrcaRouter lehnt bare Modellnamen ab. Beispiel:
   ```jsonc
   "memoryProvider": "orcarouter",

diff --git a/i18n/README.nl.md b/i18n/README.nl.md
@@ -470,7 +470,7 @@ Handmatige `memoryProvider`-modi:
 - `openai-chat`: OpenAI Chat Completions-compatibele API met tool/function calling. Dit kan werken met compatibele proxies zoals LiteLLM alleen wanneer het geselecteerde upstream-model en de proxy tool calls behouden.
 - `openai-responses`: OpenAI Responses API met function-call-output.
 - `anthropic`: Anthropic Messages API met tool use.
-- `minimax`: MiniMax Anthropic Messages-compatibel endpoint. Stel `memoryApiUrl` in op het globale endpoint (`https://api.minimax.io`) of het China-endpoint (`https://api.minimaxi.com`); het pad `/anthropic/v1/messages` en de header `x-api-key` worden automatisch toegepast. Huidige modellen zijn onder meer `MiniMax-M3` (1.000.000-token context; adaptief of uitgeschakeld thinking) en `MiniMax-M2.7` (204.800-token context; altijd-aan thinking). `MiniMax-M3` ondersteunt adaptief thinking via `memoryExtraParams`.
+- `minimax`: MiniMax Anthropic Messages-compatibel endpoint. Stel `memoryApiUrl` in op het globale endpoint (`https://api.minimax.io`) of het [China-endpoint](https://platform.minimaxi.com/docs/api-reference/text-anthropic-api.md) (`https://api.minimax.cn`); het pad `/anthropic/v1/messages` en de header `x-api-key` worden automatisch toegepast. Gebruik `MiniMax-M3.1-Flash` (aanbevolen) of `MiniMax-M3`, beide met een 1.000.000-token context. Flash denkt altijd; stel optioneel `memoryExtraParams` in op `{ "output_config": { "effort": "max" } }` met `low`, `medium`, `high`, `xhigh` of `max` (standaard). Flash weigert uitgeschakeld thinking en effort `none`, en ondersteunt geen `service_tier`. M3-thinking staat op dit endpoint standaard uit; zet het aan met `{ "thinking": { "type": "adaptive" } }`. Model-ID’s blijven expliciet geconfigureerd; bestaande custom of legacy-ID’s worden ongewijzigd doorgestuurd. Zie de [MiniMax-compatibiliteitsreferentie](https://platform.minimax.io/docs/api-reference/text-anthropic-api.md).
 - `orcarouter`: OpenAI-compatibele modelgateway met namespaced model-ID’s. `memoryApiUrl` en `memoryModel` zijn optioneel — ze defaulten naar `https://api.orcarouter.ai/v1` en `orcarouter/auto` (een routingalias die per verzoek een capabel model kiest). Als je `memoryModel` instelt, gebruik een namespaced ID zoals `openai/gpt-5.5` of `deepseek/deepseek-v4-flash`; OrcaRouter weigert bare modelnamen. Voorbeeld:
   ```jsonc
   "memoryProvider": "orcarouter",

diff --git a/i18n/README.tr.md b/i18n/README.tr.md
@@ -471,7 +471,7 @@ Manuel `memoryProvider` modları:
 - `openai-chat`: Araç/fonksiyon çağrısıyla OpenAI Chat Completions uyumlu API. Seçilen üst akış modeli ve proxy araç çağrılarını koruduğunda LiteLLM gibi uyumlu proxy’lerle çalışabilir.
 - `openai-responses`: Fonksiyon çağrısı çıktısıyla OpenAI Responses API.
 - `anthropic`: Araç kullanımıyla Anthropic Messages API.
-- `minimax`: MiniMax Anthropic Messages uyumlu uç noktası. `memoryApiUrl`’yi küresel uç noktaya (`https://api.minimax.io`) veya Çin uç noktasına (`https://api.minimaxi.com`) ayarlayın; `/anthropic/v1/messages` yolu ve `x-api-key` başlığı otomatik uygulanır. Güncel modeller arasında `MiniMax-M3` (1.000.000 jeton bağlam; uyarlanabilir veya devre dışı düşünme) ve `MiniMax-M2.7` (204.800 jeton bağlam; her zaman açık düşünme) vardır. `MiniMax-M3`, `memoryExtraParams` üzerinden uyarlanabilir düşünmeyi destekler.
+- `minimax`: MiniMax Anthropic Messages uyumlu uç noktası. `memoryApiUrl`’yi küresel uç noktaya (`https://api.minimax.io`) veya [Çin uç noktasına](https://platform.minimaxi.com/docs/api-reference/text-anthropic-api.md) (`https://api.minimax.cn`) ayarlayın; `/anthropic/v1/messages` yolu ve `x-api-key` başlığı otomatik uygulanır. `MiniMax-M3.1-Flash` (önerilen) veya `MiniMax-M3` kullanın; her ikisi de 1.000.000 jeton bağlamına sahiptir. Flash her zaman düşünür; isteğe bağlı olarak `memoryExtraParams`’ı `{ "output_config": { "effort": "max" } }` olarak `low`, `medium`, `high`, `xhigh` veya `max` (varsayılan) ile ayarlayın. Flash, devre dışı düşünmeyi ve effort `none` değerini reddeder ve `service_tier` desteklemez. Bu uç noktada M3 düşünmesi varsayılan olarak kapalıdır; `{ "thinking": { "type": "adaptive" } }` ile etkinleştirin. Model kimlikleri açıkça yapılandırılmaya devam eder; mevcut özel veya eski kimlikler değiştirilmeden iletilir. Bkz. [MiniMax uyumluluk referansı](https://platform.minimax.io/docs/api-reference/text-anthropic-api.md).
 - `orcarouter`: Ad alanlı model kimlikleriyle OpenAI uyumlu model ağ geçidi. `memoryApiUrl` ve `memoryModel` isteğe bağlıdır — varsayılanları `https://api.orcarouter.ai/v1` ve `orcarouter/auto`’dur (istek başına yetenekli bir model seçen bir yönlendirme takma adı). `memoryModel` ayarlarsanız `openai/gpt-5.5` veya `deepseek/deepseek-v4-flash` gibi ad alanlı bir kimlik kullanın; OrcaRouter çıplak model adlarını reddeder. Örnek:
   ```jsonc
   "memoryProvider": "orcarouter",

diff --git a/i18n/README.zh.md b/i18n/README.zh.md
@@ -466,7 +466,7 @@ memory({ mode: "import", inputPath: "./memories.json" });
 - `openai-chat`：兼容 OpenAI Chat Completions、带工具/函数调用的 API。仅当所选上游模型与代理保留工具调用时，这可与兼容代理（如 LiteLLM）一起工作。
 - `openai-responses`：带函数调用输出的 OpenAI Responses API。
 - `anthropic`：带工具使用的 Anthropic Messages API。
-- `minimax`：MiniMax Anthropic Messages 兼容端点。将 `memoryApiUrl` 设置为全球端点（`https://api.minimax.io`）或中国端点（`https://api.minimaxi.com`）；`/anthropic/v1/messages` 路径与 `x-api-key` 标头会自动应用。当前模型包括 `MiniMax-M3`（1,000,000-token 上下文；自适应或禁用思考）与 `MiniMax-M2.7`（204,800-token 上下文；始终开启思考）。`MiniMax-M3` 通过 `memoryExtraParams` 支持自适应思考。
+- `minimax`：MiniMax Anthropic Messages 兼容端点。将 `memoryApiUrl` 设置为全球端点（`https://api.minimax.io`）或[中国端点](https://platform.minimaxi.com/docs/api-reference/text-anthropic-api.md)（`https://api.minimax.cn`）；`/anthropic/v1/messages` 路径与 `x-api-key` 标头会自动应用。使用 `MiniMax-M3.1-Flash`（推荐）或 `MiniMax-M3`，二者均为 1,000,000-token 上下文。Flash 始终开启思考；可选将 `memoryExtraParams` 设为 `{ "output_config": { "effort": "max" } }`，取值为 `low`、`medium`、`high`、`xhigh` 或 `max`（默认）。Flash 会拒绝禁用思考与 effort `none`，且不支持 `service_tier`。在此端点上 M3 思考默认关闭；可通过 `{ "thinking": { "type": "adaptive" } }` 启用。模型 ID 仍需显式配置；已有自定义或旧版 ID 会原样透传。参见 [MiniMax 兼容性参考](https://platform.minimax.io/docs/api-reference/text-anthropic-api.md)。
 - `orcarouter`：带命名空间模型 ID 的 OpenAI 兼容模型网关。`memoryApiUrl` 与 `memoryModel` 可选 — 它们默认为 `https://api.orcarouter.ai/v1` 与 `orcarouter/auto`（按请求选择有能力模型的路由别名）。若设置 `memoryModel`，请使用命名空间 ID，例如 `openai/gpt-5.5` 或 `deepseek/deepseek-v4-flash`；OrcaRouter 拒绝裸模型名。示例：
   ```jsonc
   "memoryProvider": "orcarouter",

diff --git a/opencode-mem.example.jsonc b/opencode-mem.example.jsonc
@@ -228,14 +228,14 @@
 
   // MiniMax (Anthropic Messages-compatible endpoint, with session support):
   //   "memoryProvider": "minimax"
-  //   "memoryModel": "MiniMax-M3"                  // 1,000,000-token context
-  //   // Alternative: "MiniMax-M2.7"                // 204,800-token context; thinking always on
+  //   "memoryModel": "MiniMax-M3.1-Flash"          // 1,000,000-token context; thinking always on
+  //   // Alternative: "MiniMax-M3"                  // 1,000,000-token context; thinking off by default
   //   "memoryApiUrl": "https://api.minimax.io"        // global endpoint
   //   "memoryApiKey": "<MiniMax API key>"
-  //   // China endpoint: "memoryApiUrl": "https://api.minimaxi.com"
-  //   // MiniMax-M3 supports adaptive or disabled thinking; MiniMax-M2.7 thinking is always on.
-  //   // Optional adaptive thinking for MiniMax-M3:
-  //   "memoryExtraParams": { "thinking": { "type": "adaptive" } }
+  //   // China endpoint: "memoryApiUrl": "https://api.minimax.cn"
+  //   // Flash effort: low | medium | high | xhigh | max (default). Do not disable thinking.
+  //   "memoryExtraParams": { "output_config": { "effort": "max" } }
+  //   // For MiniMax-M3, optionally use { "thinking": { "type": "adaptive" } } instead.
 
   // OrcaRouter (OpenAI-compatible gateway, namespaced model IDs, with session support):
   //   "memoryProvider": "orcarouter"

diff --git a/src/config.ts b/src/config.ts
@@ -533,14 +533,14 @@ const CONFIG_TEMPLATE = `{
 
   // MiniMax (Anthropic Messages-compatible endpoint, with session support):
   //   "memoryProvider": "minimax"
-  //   "memoryModel": "MiniMax-M3"                  // 1,000,000-token context
-  //   // Alternative: "MiniMax-M2.7"                // 204,800-token context; thinking always on
+  //   "memoryModel": "MiniMax-M3.1-Flash"          // 1,000,000-token context; thinking always on
+  //   // Alternative: "MiniMax-M3"                  // 1,000,000-token context; thinking off by default
   //   "memoryApiUrl": "https://api.minimax.io"        // global endpoint
   //   "memoryApiKey": "<MiniMax API key>"
-  //   // China endpoint: "memoryApiUrl": "https://api.minimaxi.com"
-  //   // MiniMax-M3 supports adaptive or disabled thinking; MiniMax-M2.7 thinking is always on.
-  //   // Optional adaptive thinking for MiniMax-M3:
-  //   "memoryExtraParams": { "thinking": { "type": "adaptive" } }
+  //   // China endpoint: "memoryApiUrl": "https://api.minimax.cn"
+  //   // Flash effort: low | medium | high | xhigh | max (default). Do not disable thinking.
+  //   "memoryExtraParams": { "output_config": { "effort": "max" } }
+  //   // For MiniMax-M3, optionally use { "thinking": { "type": "adaptive" } } instead.
 
   // OrcaRouter (OpenAI-compatible gateway, namespaced model IDs, with session support):
   //   "memoryProvider": "orcarouter"

diff --git a/src/services/ai/providers/minimax.ts b/src/services/ai/providers/minimax.ts
@@ -7,16 +7,18 @@ import type { AIProviderType } from "../session/session-types.js";
  * MiniMax provider.
  *
  * MiniMax exposes an Anthropic Messages-compatible endpoint for its text
- * models. The global endpoint (https://api.minimax.io) and the China endpoint
- * (https://api.minimaxi.com) both expose the same `/anthropic/v1/messages`
- * path and authenticate with the `x-api-key` header. This provider reuses the
- * Anthropic Messages request/response handling and only overrides the resolved
- * request endpoint URL and the session provider tag, so MiniMax is
- * distinguishable from Anthropic in the session store and diagnostics.
+ * models. The global endpoint (https://api.minimax.io) and the current China
+ * endpoint (https://api.minimax.cn) both expose the same
+ * `/anthropic/v1/messages` path and authenticate with the `x-api-key` header.
+ * Configured legacy China hosts such as https://api.minimaxi.com are still
+ * resolved without remapping. This provider reuses the Anthropic Messages
+ * request/response handling and only overrides the resolved request endpoint
+ * URL and the session provider tag, so MiniMax is distinguishable from
+ * Anthropic in the session store and diagnostics.
  *
  * Users configure the base URL as `memoryApiUrl`:
  *   - global endpoint: "https://api.minimax.io"
- *   - China endpoint:  "https://api.minimaxi.com"
+ *   - China endpoint:  "https://api.minimax.cn"
  */
 export class MiniMaxProvider extends AnthropicMessagesProvider {
   constructor(config: ProviderConfig, aiSessionManager: AISessionManager) {