Thanks furas for the explanation, helped me realize the problem itself. Gonna leave a version that yields in batches and, in adjusting the batch size, you can play with the ratio the memory it uses and the response time.

@app.route("/streamed_batches")
def streamed_response_batches():
    start_time = time.time()
    mem_before = memory_usage()[0]

    BATCH_SIZE = 20

    def generate():
        yield "["
        first = True
        batch = []

        for i in range(BIG_SIZE):
            batch.append({"id": i, "value": f"Item-{i}"})

            if len(batch) >= BATCH_SIZE or i == BIG_SIZE - 1:
                # Flush this batch
                chunk = json.dumps(batch)
                if not first:
                    yield ","
                yield chunk[1:-1]
                batch = []
                first = False

        yield "]"

        mem_after = memory_usage()[0]
        elapsed = time.time() - start_time
        print(f"[STREAMED_BATCHES] Memory Before: {mem_before:.2f} MB, "
              f"After: {mem_after:.2f} MB, Elapsed: {elapsed:.2f}s")

    return Response(stream_with_context(generate()), mimetype="application/json")