A bug report and every line of Tornado—the Python web framework—went into Qwen4 Max, Alibaba's new flagship model, in a single request. No retrieval, no file selection, no tools: just 344 files, 652,519 tokens, and a concise bug description. Seventy-two seconds later, the model returned a patch that passed the project's own regression tests. The bill: 69 cents at Together AI's sale price.
This is the first clear demonstration of a 1M-token context window being used for real engineering work, not just a benchmark. The model not only fixed the bug, but also referenced precedent code elsewhere in the repository—code the bug report never mentioned, but which maintainers themselves cited in their own fix.
The Tornado test
Tornado's repository is about 340 files, including code, tests, and docs—well beyond what anyone would paste into a prompt by hand. The bug: in Tornado's websocket code, write_message() would log a noisy error when the connection closed mid-send, even though awaiting the returned Future is optional. The prompt included only the symptom, not the cause:
write_message logs "Task exception was never retrieved" when the connection closes
When a websocket connection closes while messages are still unsent (for example, the client sends its messages and a FIN, or resets the connection, without reading the replies), every write_message() call whose result was not awaited logs "Task exception was never retrieved" at ERROR level. Awaiting the Future returned by write_message is optional, so an unawaited call should not log anything. A caller that does await it should still get WebSocketClosedError.
The model traced the error to WebSocketProtocol13.write_message in tornado/websocket.py, pinpointed the uncaught exception, and wrote a patch that marks the exception as retrieved—mirroring what Tornado's BaseIOStream.write already does. None of this code was highlighted in the prompt; the model found it in the mass of files.




