Alibaba Is Paying Developers Up to 100 Million Tokens to Break Qwen3.8
Alibaba just launched the #QwenGrowthPlan — a structured community program offering up to 100 million tokens to developers who submit real-world agent task cases for Qwen3.8-Max-Preview, good or bad.
Alibaba wants your worst Qwen3.8 failures. And it’s willing to pay 100 million tokens for them.
The Qwen team launched the #QwenGrowthPlan on July 29, positioning it as a structured feedback program for Qwen3.8-Max-Preview — the 2.4-trillion-parameter model Alibaba claims trails only Fable 5 in overall capability. The program runs until August 7 and accepts submissions of real agent task cases, whether they show the model performing well or falling apart.
That last detail matters. Most developer programs reward polish. This one explicitly asks for failure.
Since launching Qwen3.8-Max-Preview, we've received valuable feedback from developers!
To help users better explore Qwen3.8's agentic capabilities, we're officially launching the #QwenGrowthPlan today! 🚀
We invite you to:
– Use Qwen3.8 to complete your real-world tasks
– Submit your good or bad cases to us
Every real task is nourishment for Qwen3.8's growth. We've prepared valuable rewards to recognize your active participation.Join us and let's grow together!
— Alibaba Qwen (@Alibaba_Qwen) July 29, 2026
What the Qwen3.8 Agent Challenge Pays
Three prize tiers structure the program:
The Featured Award goes to 10 winners and carries 100 million tokens each. Alibaba describes these as standout cases that clearly demonstrate model capabilities, with selected submissions appearing in an official promotional video.
The Excellence Award covers 50 winners at 20 million tokens each. This tier accepts both good cases and bad ones — submissions that expose issues and help the team improve the model qualify on equal footing with success stories.
The Participation Award goes to the first 10 eligible submissions by timestamp, regardless of quality. Each receives 10 million tokens.
How to Submit
Submissions go to QwenOfficial@alibabacloud.com. The subject line must read #QwenGrowthPlan followed by a task description. The body needs four things: the agent task scenario, the exact prompt used, Qwen’s actual output, and the developer’s own evaluation or suggestions. Screenshots, code files, and compressed archives are all acceptable attachments.
Alibaba is also accepting submissions directly on X by tagging @Alibaba_Qwen with the hashtag.
Why Alibaba Wants the Bad Cases
Qwen3.8-Max-Preview launched with unusually thin benchmark documentation. The 2.4T-parameter model went live in preview before Alibaba published the standard suite of evaluation scores most frontier releases carry. That gap made independent developer testing the primary signal available to the community.
The Growth Plan formalises what was already happening organically. Developers running real agentic workflows on the model were already posting results on X. Now Alibaba is incentivising those developers to structure and submit what they find, rather than just tweet about it.
The focus on agent tasks is specific. Qwen3.8-Max is marketed as an agentic model — one built for multi-step task execution rather than single-turn question answering. Real-world agent pipelines expose failure modes that synthetic benchmarks routinely miss: tool call loops, instruction drift across long contexts, incorrect state tracking. These are exactly the cases worth 20 million tokens to Alibaba’s team.
The Token Math
At Qwen’s current public pricing, 100 million tokens represents a substantial API credit. For developers running production agentic workloads, it’s months of compute. The Participation Award alone — 10 million tokens for simply being among the first 10 to submit — covers the kind of extended testing that would otherwise cost real money.
Alibaba has been aggressive on pricing across the Qwen family, using deep discounts on API access to drive adoption against competitors. The Growth Plan extends that approach into community feedback: rather than buying evaluations from a benchmark lab, the team is purchasing real-world signal directly from the developers most likely to stress-test the model in ways that matter.
The deadline is August 7. Submissions open now at QwenOfficial@alibabacloud.com.
Frequently Asked Questions
What is the #QwenGrowthPlan?
It’s a structured community feedback program from Alibaba’s Qwen team. Developers submit real-world agent task cases for Qwen3.8-Max-Preview — good or bad — and are rewarded with token prizes based on the quality and timing of submissions.
How do I enter the Qwen3.8 agent challenge?
Email QwenOfficial@alibabacloud.com with the subject line #QwenGrowthPlan plus a task description. Include the agent scenario, the exact prompt, Qwen’s actual response, and your evaluation. Screenshots and code files are accepted as attachments. You can also post directly on X and tag @Alibaba_Qwen.
What counts as a valid submission?
Any real agent task completed using Qwen3.8. The Excellence Award tier explicitly accepts cases where the model fails or underperforms — bug reports and failure documentation are as valid as success stories.
When is the deadline?
August 7, 2026. The Participation Award goes to the first 10 eligible submissions by timestamp, so early entry has a direct advantage for that tier.
