diff --git a/apps/gateway/src/chat/tools/parse-provider-response.spec.ts b/apps/gateway/src/chat/tools/parse-provider-response.spec.ts
index 94a4570cc1..8ded2810c3 100644
--- a/apps/gateway/src/chat/tools/parse-provider-response.spec.ts
+++ b/apps/gateway/src/chat/tools/parse-provider-response.spec.ts
@@ -824,6 +824,68 @@ describe("parseProviderResponse", () => {
 
 			expect(result.finishReason).toBe("upstream_error");
 		});
+
+		it("maps 'end_turn' finish reason to 'stop' for groq", () => {
+			const json = {
+				choices: [
+					{
+						message: { content: "Hello", role: "assistant" },
+						finish_reason: "end_turn",
+					},
+				],
+				usage: {
+					prompt_tokens: 10,
+					completion_tokens: 5,
+					total_tokens: 15,
+				},
+			};
+
+			const result = parseProviderResponse(
+				"groq",
+				"llama-3.3-70b-versatile",
+				json,
+			);
+
+			expect(result.finishReason).toBe("stop");
+		});
+
+		it("maps 'tool_use' finish reason to 'tool_calls' for together-ai", () => {
+			const json = {
+				choices: [
+					{
+						message: {
+							role: "assistant",
+							content: null,
+							tool_calls: [
+								{
+									id: "call_1",
+									type: "function",
+									function: {
+										name: "get_weather",
+										arguments: '{"city":"San Francisco"}',
+									},
+								},
+							],
+						},
+						finish_reason: "tool_use",
+					},
+				],
+				usage: {
+					prompt_tokens: 10,
+					completion_tokens: 5,
+					total_tokens: 15,
+				},
+			};
+
+			const result = parseProviderResponse(
+				"together-ai",
+				"deepseek-ai/DeepSeek-V3",
+				json,
+			);
+
+			expect(result.finishReason).toBe("tool_calls");
+			expect(result.toolResults).toHaveLength(1);
+		});
 	});
 
 	describe("refusal finish reason", () => {
diff --git a/apps/gateway/src/chat/tools/parse-provider-response.ts b/apps/gateway/src/chat/tools/parse-provider-response.ts
index 33b92c60b0..b721f0e72f 100644
--- a/apps/gateway/src/chat/tools/parse-provider-response.ts
+++ b/apps/gateway/src/chat/tools/parse-provider-response.ts
@@ -1074,7 +1074,10 @@ export function parseProviderResponse(
 				reasoningContent = hasReasoning ? aggregatedReasoning : null;
 				finishReason = allChoices[0]?.finish_reason ?? null;
 
-				if (finishReason === "abort") {
+				// Map non-standard finish reasons to OpenAI-compatible values
+				if (finishReason === "end_turn") {
+					finishReason = "stop";
+				} else if (finishReason === "abort") {
 					logger.warn("Upstream sent abort finish_reason", {
 						provider: usedProvider,
 						model: usedModel,
@@ -1087,6 +1090,8 @@ export function parseProviderResponse(
 					// "abort" is an upstream-initiated interruption, not a client
 					// cancellation, so it counts as an upstream error.
 					finishReason = "upstream_error";
+				} else if (finishReason === "tool_use") {
+					finishReason = "tool_calls";
 				}
 
 				// ZAI-specific fix for incorrect finish_reason in tool response scenarios