diff --git a/src/agents/model-ref-profile.test.ts b/src/agents/model-ref-profile.test.ts
index f85c7d37675..f1eccdc431a 100644
--- a/src/agents/model-ref-profile.test.ts
+++ b/src/agents/model-ref-profile.test.ts
@@ -66,4 +66,23 @@ describe("splitTrailingAuthProfile", () => {
       profile: "work",
     });
   });
+
+  it("keeps @q* quant suffixes in model ids", () => {
+    expect(splitTrailingAuthProfile("lmstudio-mb-pro/gemma-4-31b-it@q8_0")).toEqual({
+      model: "lmstudio-mb-pro/gemma-4-31b-it@q8_0",
+    });
+  });
+
+  it("supports auth profiles after @q* quant suffixes", () => {
+    expect(splitTrailingAuthProfile("lmstudio-mb-pro/gemma-4-31b-it@q8_0@work")).toEqual({
+      model: "lmstudio-mb-pro/gemma-4-31b-it@q8_0",
+      profile: "work",
+    });
+  });
+
+  it("keeps @4bit/@8bit quant suffixes in model ids", () => {
+    expect(splitTrailingAuthProfile("lmstudio-mb-pro/gemma-4-31b@4bit")).toEqual({
+      model: "lmstudio-mb-pro/gemma-4-31b@4bit",
+    });
+  });
 });
diff --git a/src/agents/model-ref-profile.ts b/src/agents/model-ref-profile.ts
index 7437a554590..519874e69d6 100644
--- a/src/agents/model-ref-profile.ts
+++ b/src/agents/model-ref-profile.ts
@@ -14,6 +14,9 @@ export function splitTrailingAuthProfile(raw: string): {
   }
 
   const versionSuffix = trimmed.slice(profileDelimiter + 1);
+
+  // Keep well-known "version" suffixes (ex: @20251001) as part of the model id,
+  // but allow an auth profile suffix *after* them (ex: ...@20251001@work).
   if (/^\d{8}(?:@|$)/.test(versionSuffix)) {
     const nextDelimiter = trimmed.indexOf("@", profileDelimiter + 9);
     if (nextDelimiter < 0) {
@@ -22,6 +25,20 @@ export function splitTrailingAuthProfile(raw: string): {
     profileDelimiter = nextDelimiter;
   }
 
+  // Keep local model quant suffixes (common in LM Studio/Ollama catalogs) as part
+  // of the model id. These often use '@' (ex: gemma-4-31b-it@q8_0) which would
+  // otherwise be misinterpreted as an auth profile delimiter.
+  //
+  // If an auth profile is needed, it can still be specified as a second suffix:
+  //   lmstudio/foo@q8_0@work
+  if (/^(?:q\d+(?:_[a-z0-9]+)*|\d+bit)(?:@|$)/i.test(versionSuffix)) {
+    const nextDelimiter = trimmed.indexOf("@", profileDelimiter + 1);
+    if (nextDelimiter < 0) {
+      return { model: trimmed };
+    }
+    profileDelimiter = nextDelimiter;
+  }
+
   const model = trimmed.slice(0, profileDelimiter).trim();
   const profile = trimmed.slice(profileDelimiter + 1).trim();
   if (!model || !profile) {