fixes ocr for newer open ai models

src/app/api/labs/ocr/extract/__tests__/route.test.ts

@@ -136,6 +136,16 @@ describe("POST /api/labs/ocr/extract — text mode budget", () => {
       "2026-06-26",
     );
   });
+
+  it("identifies provider failures instead of blaming image quality", async () => {
+    vi.mocked(runOcrExtraction).mockRejectedValue(new Error("provider down"));
+
+    const res = await POST(textReq());
+    const body = (await res.json()) as { error: string };
+
+    expect(res.status).toBe(502);
+    expect(body.error).toContain("configured AI provider");
+  });
 });
 
 describe("POST /api/labs/ocr/extract — vision PDF rasterization", () => {

src/app/api/labs/ocr/extract/route.ts

@@ -63,6 +63,25 @@ const EXTRACT_WINDOW_MS = 60 * 60 * 1000;
 /** OCR'd text is bounded in the schema; cap the JSON body proportionally. */
 const TEXT_BODY_MAX_BYTES = 512 * 1024;
 
+function providerFailureMeta(error: unknown, mode?: "text") {
+  const err = error as {
+    httpStatus?: unknown;
+    model?: unknown;
+    bodyExcerpt?: unknown;
+  };
+  return {
+    reason: "provider_error",
+    ...(mode ? { mode } : {}),
+    ...(typeof err.httpStatus === "number"
+      ? { upstreamStatus: err.httpStatus }
+      : {}),
+    ...(typeof err.model === "string" ? { model: err.model } : {}),
+    ...(typeof err.bodyExcerpt === "string"
+      ? { upstreamError: err.bodyExcerpt }
+      : {}),
+  };
+}
+
 export const POST = apiHandler(async (request: Request) => {
   const { user } = await requireAuth();
 
@@ -191,11 +210,13 @@ async function handleTextExtract(
     }
     annotate({
       action: { name: "labs.ocr.extractFailed" },
-      meta: { reason: "provider_error", mode: "text" },
-    });
-    return apiError("Couldn't read the report. Try a clearer photo.", 502, {
-      errorCode: "labs.ocr.extractFailed",
+      meta: providerFailureMeta(err, "text"),
     });
+    return apiError(
+      "The configured AI provider could not process this report. Check the provider configuration and retry.",
+      502,
+      { errorCode: "labs.ocr.extractFailed" },
+    );
   }
 }
 
@@ -397,11 +418,13 @@ async function handleVisionExtract(
       }
       annotate({
         action: { name: "labs.ocr.extractFailed" },
-        meta: { reason: "provider_error" },
-      });
-      return apiError("Couldn't read the report. Try a clearer photo.", 502, {
-        errorCode: "labs.ocr.extractFailed",
+        meta: providerFailureMeta(err),
       });
+      return apiError(
+        "The configured AI provider could not process this report. Check the provider configuration and retry.",
+        502,
+        { errorCode: "labs.ocr.extractFailed" },
+      );
     }
   } catch (err) {
     // A guard threw after the reservation (e.g. consent races) — refund fully.

src/lib/ai/__tests__/vision-capability.test.ts

@@ -37,6 +37,9 @@ describe("supportsVisionForConfig", () => {
     ] as const) {
       expect(supportsVisionForConfig(providerType, "gpt-4o")).toBe(true);
       expect(supportsVisionForConfig(providerType, "gpt-4o-mini")).toBe(true);
+      expect(supportsVisionForConfig(providerType, "gpt-5.6-terra")).toBe(
+        true,
+      );
       expect(supportsVisionForConfig(providerType, "gpt-4.1")).toBe(true);
       expect(supportsVisionForConfig(providerType, "gpt-4-turbo")).toBe(true);
       // Full o-series reasoning models read images.

src/lib/ai/ai-budgets.ts

@@ -189,13 +189,14 @@ export const AI_BUDGETS = {
    * v1.18.9 — Lab-OCR structured extraction. ONE vision call transcribes a
    * photo / PDF of a paper lab report into a JSON array of analyte rows
    * (analyte + value/valueText + unit + reference range + date + per-field
-   * confidence). A full panel is a dozen-plus rows of compact JSON; 4000
-   * tokens covers a dense report plus the envelope. Temperature 0 — the task
+   * confidence). A full panel is a dozen-plus rows of compact JSON; 8000
+   * tokens leaves room for GPT-5 reasoning plus the dense report envelope.
+   * Temperature 0 — the task
    * is faithful transcription, not generation, so the lowest setting minimises
    * invented values. Vision input tokens are expensive, so the route also
    * gates with a tight 6/hour rate bucket + reserveBudget + upload-size cap.
    */
-  ocrExtract: { temperature: 0, maxTokens: 4000 },
+  ocrExtract: { temperature: 0, maxTokens: 8000 },
 
   /**
    * v1.20.1 — Lab-OCR text-mode structuring. The browser OCR's the image

src/lib/ai/vision-capability.ts

@@ -63,6 +63,7 @@ function isVisionReasoningSlug(m: string): boolean {
 function openaiSupportsVision(model: string): boolean {
   const m = model.toLowerCase();
   return (
+    m.startsWith("gpt-5") ||
     m.startsWith("gpt-4o") ||
     m.startsWith("gpt-4.1") ||
     m.startsWith("gpt-4-turbo") ||