diff --git a/src/core/api/providers/cerebras.ts b/src/core/api/providers/cerebras.ts
index 23881f7525..921371e1d4 100644
--- a/src/core/api/providers/cerebras.ts
+++ b/src/core/api/providers/cerebras.ts
@@ -192,7 +192,18 @@ export class CerebrasHandler implements ApiHandler {
 		} catch (error: any) {
 			// Enhanced error handling for Cerebras API
 			if (error?.status === 429 || error?.code === "rate_limit_exceeded") {
-				// Rate limit error - will be handled by retry decorator with patient backoff
+				// Check if this is actually a context window error disguised as rate limit
+				// Cerebras rate limiter counts input tokens, so oversized context can trigger 429
+				const errorMessage = String(error?.message || "")
+				const isContextRelated = /reduce.*length|context.*length|too.*many.*tokens|tokens.*exceed|input.*too.*long/i.test(
+					errorMessage,
+				)
+				if (isContextRelated) {
+					// Preserve original error so context window detection can catch it
+					error.status = 400 // Treat as bad request for context window detection
+					throw error
+				}
+				// Regular rate limit error - will be handled by retry decorator with patient backoff
 				const _limits = this.getRateLimits()
 				throw new Error(`Cerebras API rate limit exceeded.`)
 			} else if (error?.status === 401) {
@@ -203,8 +214,9 @@ export class CerebrasHandler implements ApiHandler {
 				// Server errors - retryable
 				throw new Error(`Cerebras API server error (${error.status}): ${error.message || "Unknown server error"}`)
 			} else if (error?.status === 400) {
-				// Client errors - not retryable
-				throw new Error(`Cerebras API bad request: ${error.message || "Invalid request parameters"}`)
+				// Client errors (including context window exceeded) - preserve original error
+				// so checkIsCerebrasContextWindowError can detect it and trigger compaction
+				throw error
 			}
 
 			// Re-throw original error for other cases
diff --git a/src/core/context/context-management/context-error-handling.ts b/src/core/context/context-management/context-error-handling.ts
index f51919652f..5b5c05fb58 100644
--- a/src/core/context/context-management/context-error-handling.ts
+++ b/src/core/context/context-management/context-error-handling.ts
@@ -67,7 +67,21 @@ function checkIsCerebrasContextWindowError(response: any): boolean {
 		const status = response?.status ?? response?.code ?? response?.error?.status ?? response?.response?.status
 		const message: string = String(response?.message || response?.error?.message || "")
 
-		return String(status) === "400" && message.includes("Please reduce the length of the messages or completion")
+		if (String(status) !== "400") {
+			return false
+		}
+
+		// Known Cerebras context window error patterns
+		const CEREBRAS_CONTEXT_PATTERNS = [
+			/reduce.*length.*messages.*completion/i,
+			/\bcontext\s*(?:length|window)\b/i,
+			/\bmaximum\s*context\b/i,
+			/\b(?:input\s*)?tokens?\s*exceed/i,
+			/\btoo\s*many\s*tokens?\b/i,
+			/input is too long/i,
+		] as const
+
+		return CEREBRAS_CONTEXT_PATTERNS.some((pattern) => pattern.test(message))
 	} catch {
 		return false
 	}
diff --git a/src/utils/model-utils.ts b/src/utils/model-utils.ts
index 9e94dcc606..f3034c0529 100644
--- a/src/utils/model-utils.ts
+++ b/src/utils/model-utils.ts
@@ -97,6 +97,8 @@ export function isGLMModelFamily(id: string): boolean {
 	return (
 		modelId.includes("glm-4.6") ||
 		modelId.includes("glm-4.5") ||
+		modelId.includes("glm-4.7") ||
+		modelId.includes("zai-glm") || // Cerebras pattern (e.g., zai-glm-4.7)
 		modelId.includes("z-ai/glm") ||
 		modelId.includes("zai-org/glm")
 	)
@@ -163,7 +165,8 @@ export function isNextGenModelFamily(id: string): boolean {
 		isGemini3ModelFamily(modelId) ||
 		isNextGenOpenSourceModelFamily(modelId) ||
 		isDeepSeek32ModelFamily(modelId) ||
-		isDeepSeekNativeModelFamily(modelId)
+		isDeepSeekNativeModelFamily(modelId) ||
+		isGLMModelFamily(modelId)
 	)
 }