diff --git a/CHANGELOG.md b/CHANGELOG.md
index 5a42442..912875a 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -2,6 +2,8 @@
 
 ## Unreleased
 
+- Updated managed GonkaGate `limit.context` values to match the current
+  deployed windows: `240000` for Kimi/Qwen and `180000` for MiniMax.
 - Wrote the validated GonkaGate chat-completions catalog into the managed
   provider config so Kilo's OpenCode-style `/models` picker can switch between
   Kimi K2.6, Qwen3 235B A22B Instruct 2507 FP8, and MiniMax M2.7.
diff --git a/docs/gonkagate-x-kilo.md b/docs/gonkagate-x-kilo.md
index 5121748..ad9db1d 100644
--- a/docs/gonkagate-x-kilo.md
+++ b/docs/gonkagate-x-kilo.md
@@ -135,7 +135,7 @@ Right now the public default is deliberately small:
   `qwen/qwen3-235b-a22b-instruct-2507-fp8`, and
   `minimaxai/minimax-m2.7`
 - managed limits: `limit.output = 8192` for all validated entries;
-  `limit.context = 262144` for Kimi/Qwen and `204800` for MiniMax
+  `limit.context = 240000` for Kimi/Qwen and `180000` for MiniMax
 
 We are treating model support as a curated list, not as a vague "it probably
 works" promise.
diff --git a/docs/how-it-works.md b/docs/how-it-works.md
index 4fac58a..c3d0724 100644
--- a/docs/how-it-works.md
+++ b/docs/how-it-works.md
@@ -30,7 +30,7 @@ Current public limit:
   `@kilocode/cli >=7.2.0`, `chat/completions`, and non-Windows production
   claims
 - the curated default is
-  `moonshotai/Kimi-K2.6` with `limit.context = 262144` and
+  `moonshotai/Kimi-K2.6` with `limit.context = 240000` and
   `limit.output = 8192`
 - the written provider config includes the validated chat-completions model
   catalog, currently `moonshotai/Kimi-K2.6`,
diff --git a/docs/release-readiness.md b/docs/release-readiness.md
index 030126d..21db173 100644
--- a/docs/release-readiness.md
+++ b/docs/release-readiness.md
@@ -24,8 +24,8 @@ docs, and tests for these facts:
   `qwen/qwen3-235b-a22b-instruct-2507-fp8`, and
   `minimaxai/minimax-m2.7`
 - curated model limits for the validated catalog:
-  `limit.output = 8192` for all validated entries; `limit.context = 262144`
-  for Kimi/Qwen and `204800` for MiniMax
+  `limit.output = 8192` for all validated entries; `limit.context = 240000`
+  for Kimi/Qwen and `180000` for MiniMax
 - managed secret path: `~/.gonkagate/kilo/api-key`
 - project scope stays secret-free and still depends on a compatible
   user-level `provider.gonkagate` definition on each machine
@@ -34,24 +34,21 @@ docs, and tests for these facts:
 
 ## External Evidence Captured
 
-The current default-model contract is backed by product direction plus public
-Moonshot metadata checked on 2026-04-29:
+The current model-limit contract is backed by product direction plus Gonka
+deployment metadata checked on 2026-06-23:
 
-- the Kimi K2.6 model card lists `moonshotai/Kimi-K2.6` and documents a 256K
-  context window
+- Gonka deployment model args list Kimi K2.6 and Qwen3 235B A22B Instruct 2507
+  FP8 with `--max-model-len 240000`
+- Gonka deployment model args list `minimaxai/minimax-m2.7` with
+  `--max-model-len 180000`
 - the model card documents OpenAI-compatible chat completions access through
   Moonshot's API
 - the API docs describe the supported request path as `POST /v1/chat/completions`
 - the package writes `limit.output = 8192` for validated catalog entries as
   the installer-managed Kilo compatibility clamp because Kilo `7.2.0` requires
   a numeric output limit in custom model config
-- the April 14, 2026 Kilo compatibility spike captured
-  `qwen/qwen3-235b-a22b-instruct-2507-fp8` as a validated
-  chat-completions model with a 262K context window and `maxTokens` 8192
-- NVIDIA NIM metadata checked on 2026-05-29 lists
-  `minimaxai/minimax-m2.7` as MiniMax M2.7 with a 204,800 input context
-  length; the package keeps the same installer-managed `limit.output = 8192`
-  compatibility clamp for this catalog entry
+- the package keeps the same installer-managed `limit.output = 8192`
+  compatibility clamp for every validated catalog entry
 - npm registry metadata checked on 2026-04-29 showed `@kilocode/cli` patch
   releases in the `7.2.x` line, including `7.2.14`, with both `kilo` and
   `kilocode` binaries still exposed by the wrapper package
diff --git a/docs/specs/kilo-setup-prd/compatibility-spike-notes.md b/docs/specs/kilo-setup-prd/compatibility-spike-notes.md
index cf1685c..4abbb69 100644
--- a/docs/specs/kilo-setup-prd/compatibility-spike-notes.md
+++ b/docs/specs/kilo-setup-prd/compatibility-spike-notes.md
@@ -23,7 +23,7 @@ not contain raw `kilo debug config` output or real secrets.
   - example model id:
     `qwen/qwen3-235b-a22b-instruct-2507-fp8`
 - GonkaGate public model page checked:
-  - `qwen/qwen3-235b-a22b-instruct-2507-fp8` context window: `262K`
+  - `qwen/qwen3-235b-a22b-instruct-2507-fp8` context window: `240000`
   - GonkaGate OpenClaw custom-provider guidance documents `maxTokens: 8192`
     for the same model.
 
diff --git a/docs/specs/kilo-setup-prd/spec.md b/docs/specs/kilo-setup-prd/spec.md
index c96c18a..524dd60 100644
--- a/docs/specs/kilo-setup-prd/spec.md
+++ b/docs/specs/kilo-setup-prd/spec.md
@@ -66,8 +66,8 @@ Redacted evidence notes are preserved in
 - A full `.../chat/completions` URL must not be written to any Kilo base URL
   field; Kilo appends `chat/completions` through the OpenAI-compatible adapter.
 - Production setup should leave `small_model` untouched by default.
-- GonkaGate public sources prove the selected model's context window as `262K`
-  tokens, interpreted for Kilo as `262144`.
+- GonkaGate deployment sources prove the selected model's context window as
+  `240000` tokens.
 - GonkaGate public OpenClaw guidance documents `maxTokens: 8192` for the
   selected model, which the installer writes as Kilo `limit.output`.
 - No safe `GONKAGATE_API_KEY` was present during the spike, so live GonkaGate
@@ -538,7 +538,7 @@ Expected production managed provider shape:
           "name": "Qwen3 235B A22B Instruct 2507 FP8",
           "tool_call": true,
           "limit": {
-            "context": 262144,
+            "context": 240000,
             "output": "BLOCKED_UNTIL_GONKAGATE_LIMIT_IS_PROVEN",
           },
         },
@@ -576,8 +576,8 @@ Evidence:
 - The same fake smoke proved that writing a full `.../chat/completions` URL
   produces a doubled `/chat/completions/chat/completions` path.
 
-The `limit.context` value is proven from the GonkaGate public model page as
-`262K`, mapped to the integer `262144`. GonkaGate OpenClaw guidance documents
+The `limit.context` value is proven from the GonkaGate deployment metadata as
+`240000`. GonkaGate OpenClaw guidance documents
 `maxTokens: 8192` for the same model, so the installer writes
 `limit.output = 8192`. Kilo docs state that `context: 0` disables compaction and
 context-size-dependent usage tracking, while `output: 0` falls back to Kilo's
diff --git a/docs/user-guide.md b/docs/user-guide.md
index 7eb553b..6336fae 100644
--- a/docs/user-guide.md
+++ b/docs/user-guide.md
@@ -30,7 +30,7 @@ Today the public package is intentionally narrow:
   `qwen/qwen3-235b-a22b-instruct-2507-fp8`, and
   `minimaxai/minimax-m2.7`
 - managed model limits: `limit.output = 8192` for all validated entries;
-  `limit.context = 262144` for Kimi/Qwen and `204800` for MiniMax
+  `limit.context = 240000` for Kimi/Qwen and `180000` for MiniMax
 - no native Windows production claim yet
 
 ## Before You Run It
diff --git a/src/constants/models.ts b/src/constants/models.ts
index 1ebf5c6..55dab3e 100644
--- a/src/constants/models.ts
+++ b/src/constants/models.ts
@@ -84,14 +84,14 @@ export const CURATED_MODEL_REGISTRY = Object.freeze({
     adapterPackage: "@ai-sdk/openai-compatible",
     displayName: "Kimi K2.6",
     limits: {
-      context: 262144,
+      context: 240000,
       output: 8192,
     },
     modelId: "moonshotai/Kimi-K2.6",
     recommended: true,
     runtimeCompatibility: {
       notes: [
-        "Moonshot model metadata lists Kimi K2.6 with a 256K context window and OpenAI-compatible chat completions access.",
+        "Gonka deployment metadata lists Kimi K2.6 with a 240,000 token context window and OpenAI-compatible chat completions access.",
         "The installer writes limit.output = 8192 as a conservative Kilo 7.2.0 compatibility clamp for this default.",
       ],
     },
@@ -102,14 +102,14 @@ export const CURATED_MODEL_REGISTRY = Object.freeze({
     adapterPackage: "@ai-sdk/openai-compatible",
     displayName: "Qwen3 235B A22B Instruct 2507 FP8",
     limits: {
-      context: 262144,
+      context: 240000,
       output: 8192,
     },
     modelId: "qwen/qwen3-235b-a22b-instruct-2507-fp8",
     recommended: false,
     runtimeCompatibility: {
       notes: [
-        "Official GonkaGate model metadata lists this model as available with a 262K context window.",
+        "Gonka deployment metadata lists this model as available with a 240,000 token context window.",
         "Official GonkaGate OpenClaw provider guidance lists this model with maxTokens 8192.",
         "Kilo 7.2.0 rejects custom model entries without a numeric limit.output value, so the installer writes limit.output = 8192 for this validated default.",
       ],
@@ -121,14 +121,14 @@ export const CURATED_MODEL_REGISTRY = Object.freeze({
     adapterPackage: "@ai-sdk/openai-compatible",
     displayName: "MiniMax M2.7",
     limits: {
-      context: 204800,
+      context: 180000,
       output: 8192,
     },
     modelId: "minimaxai/minimax-m2.7",
     recommended: false,
     runtimeCompatibility: {
       notes: [
-        "NVIDIA NIM metadata lists minimaxai/minimax-m2.7 as MiniMax M2.7 with a 204,800 input context length.",
+        "Gonka deployment metadata lists minimaxai/minimax-m2.7 as MiniMax M2.7 with a 180,000 token context window.",
         "The installer writes limit.output = 8192 as the existing conservative Kilo 7.2.0 compatibility clamp for validated catalog entries.",
       ],
     },
diff --git a/test/cli.test.ts b/test/cli.test.ts
index f4f1d66..840018b 100644
--- a/test/cli.test.ts
+++ b/test/cli.test.ts
@@ -27,7 +27,7 @@ const VALIDATED_MODEL: RecommendedProductionDefaultCuratedModel = {
   displayName: "Kimi K2.6",
   key: MODEL_KEY,
   limits: {
-    context: 262144,
+    context: 240000,
     output: 8192,
   },
   modelId: "moonshotai/Kimi-K2.6",
diff --git a/test/install/models.test.ts b/test/install/models.test.ts
index f1b26b5..fdbbab4 100644
--- a/test/install/models.test.ts
+++ b/test/install/models.test.ts
@@ -12,7 +12,7 @@ test("the default curated model is the shipped validated Kimi public default", (
 
   assert.equal(model.adapterPackage, "@ai-sdk/openai-compatible");
   assert.equal(model.validationStatus, "validated");
-  assert.equal(model.limits?.context, 262144);
+  assert.equal(model.limits?.context, 240000);
   assert.equal(model.limits?.output, 8192);
   assert.deepEqual(getValidatedModelKeys(), [
     "kimi-k2.6",
diff --git a/test/install/test-model-catalog.ts b/test/install/test-model-catalog.ts
index 10beeba..56fa01d 100644
--- a/test/install/test-model-catalog.ts
+++ b/test/install/test-model-catalog.ts
@@ -6,7 +6,7 @@ export const TEST_VALIDATED_MODEL = {
   displayName: "Kimi K2.6",
   key: "kimi-k2.6",
   limits: {
-    context: 262144,
+    context: 240000,
     output: 8192,
   },
   modelId: "moonshotai/Kimi-K2.6",
diff --git a/test/package-contract.test.ts b/test/package-contract.test.ts
index bb0c39d..fc0ddc3 100644
--- a/test/package-contract.test.ts
+++ b/test/package-contract.test.ts
@@ -138,14 +138,14 @@ test("curated model registry exposes Kimi as the shipped validated default", ()
   const minimax = CURATED_MODEL_REGISTRY["minimax-m2.7"];
 
   assert.equal(kimi.adapterPackage, "@ai-sdk/openai-compatible");
-  assert.deepEqual(kimi.limits, { context: 262144, output: 8192 });
+  assert.deepEqual(kimi.limits, { context: 240000, output: 8192 });
   assert.equal(kimi.modelId, "moonshotai/Kimi-K2.6");
   assert.equal(kimi.recommended, true);
   assert.equal(kimi.transport, "chat_completions");
   assert.equal(kimi.validationStatus, "validated");
 
   assert.equal(qwen.adapterPackage, "@ai-sdk/openai-compatible");
-  assert.equal(qwen.limits?.context, 262144);
+  assert.equal(qwen.limits?.context, 240000);
   assert.equal(qwen.limits?.output, 8192);
   assert.equal(qwen.modelId, "qwen/qwen3-235b-a22b-instruct-2507-fp8");
   assert.equal(qwen.recommended, false);
@@ -153,7 +153,7 @@ test("curated model registry exposes Kimi as the shipped validated default", ()
   assert.equal(qwen.validationStatus, "validated");
 
   assert.equal(minimax.adapterPackage, "@ai-sdk/openai-compatible");
-  assert.equal(minimax.limits?.context, 204800);
+  assert.equal(minimax.limits?.context, 180000);
   assert.equal(minimax.limits?.output, 8192);
   assert.equal(minimax.modelId, "minimaxai/minimax-m2.7");
   assert.equal(minimax.recommended, false);
@@ -172,9 +172,9 @@ test("managed provider config includes the required Kilo model limits", () => {
     models: Record<string, { limit?: { context?: number; output?: number } }>;
   };
 
-  assert.equal(providerConfig.models["kimi-k2.6"]?.limit?.context, 262144);
+  assert.equal(providerConfig.models["kimi-k2.6"]?.limit?.context, 240000);
   assert.equal(providerConfig.models["kimi-k2.6"]?.limit?.output, 8192);
-  assert.equal(providerConfig.models["minimax-m2.7"]?.limit?.context, 204800);
+  assert.equal(providerConfig.models["minimax-m2.7"]?.limit?.context, 180000);
   assert.equal(providerConfig.models["minimax-m2.7"]?.limit?.output, 8192);
 });