AIToday
Large Language ModelsAI Coding AssistantsTogether AI BlogPublished: Nov 22, 2026, 06:01 JST

Qwen4 Max fixes Tornado bug from 652,519-token repo for 69 cents

Send on LINE
Qwen4 Max fixes Tornado bug from 652,519-token repo for 69 cents

3 Key Points

  1. What happened

    Together AI sent Qwen4 Max a bug report plus all 344 files of Tornado (652,519 tokens) with no tools or retrieval; in 72 seconds it returned a patch that passed the maintainers' two regression tests, for 69 cents.

  2. Why it matters

    The model also pointed to precedent in tornado/iostream.py that the bug report never mentioned, and Tornado's maintainers cite the same code in their fix — suggesting whole-repository coding can work in one call without hand-picked files.

WHO IT HITSDevelopers and engineering teams working on large codebases get a concrete cost and speed benchmark for whole-repository bug fixing, while teams that send entire codebases in prompts may need to check for secrets first, as Together AI warns.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Context & Analysis

Together AI picked a fix Tornado's maintainers committed on November 18, 2026 — two days after Alibaba released Qwen4 Max on November 16 — and tested against the repository as it stood just before that commit. That timing is what makes the result a genuine forward test rather than a replay of code the model might have seen.

The details of the run are what stand out. The model had no tools, no file selection and no retrieval step; all 344 files went in as one prompt. Its explanation reached for BaseIOStream.write in tornado/iostream.py, a file the bug report never names, and the maintainers' own commit makes the same comparison. Together AI then ran Tornado's full suite of 1,673 tests with the patch applied, and none failed.

Cost behavior is the other half of the story. The first request cost about 69 cents at the sale price, nearly all of it input tokens, or 98 cents at list price. Repeating a roughly 630,000-token request came back almost entirely from Together AI's automatic prompt cache: the first token arrived in 8 seconds instead of 54, and input cost about 3 cents instead of 66. Alibaba, for its part, positions Qwen4 Max — a sparse mixture-of-experts model with 2.4 trillion parameters, 71 billion active per token — for long agent work such as coding across large repositories.

FAQ
How much does Qwen4 Max cost on Together AI?
List price is $1.50 per 1M input tokens, $4.50 per 1M output tokens and $0.075 per 1M cached input tokens. Through November 22, 2026 it is 30% off: $1.05, $3.15 and $0.0525.
How large a codebase can Qwen4 Max handle?
It has a 1M-token context window on Together AI. In the test, a request with 652,519 input tokens returned its first token after 54 seconds.
Can I turn off thinking to save tokens?
Yes. Pass reasoning={"enabled": False} in the Python SDK or "reasoning": {"enabled": false} in the request body; thinking is on by default and its tokens count as output tokens.
Together AI BlogRead Original Article
Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleGyro Governance Lab releases AIR Office for AI safety research