📊 Full opportunity report: Is Qwen3.8-Max The AI Second Place? The Numbers Tell A Complex Story on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Alibaba has publicly released details of Qwen3.8-Max, claiming it as the second-largest AI model after Fable 5. Its benchmark results show strong performance, but only on selected tasks. The open-weight release next week will clarify its practical impact.
Alibaba has confirmed the release of Qwen3.8-Max, a 2.4 trillion-parameter AI model, with full benchmark results published today. The company claims it is second only to GPT-5.6 in performance, marking a significant milestone in large-scale AI development. This announcement, following weeks of speculation, provides the first detailed look at the model’s capabilities and specifications, and the open weights will be available next week.
Alibaba’s Qwen3.8-Max features approximately 95 billion active parameters per query within its 2.4 trillion total parameters, utilizing a sparse mixture-of-experts architecture based on Qwen3.5. The model is multimodal, supporting text, images, and video inputs, with text output. The benchmark table, published today, shows the model achieving top scores in several tasks, such as Terminal-Bench 2.1 (86.6), outperforming Claude models and only trailing GPT-5.6 Sol at 88.8. It also leads in paper-based benchmarks like PaperBench at 93.0 and demonstrates strong agentic capabilities, notably improving long-horizon task performance, such as in DeepSWE, which jumped from 21.6 to 56.6.
However, the model trails significantly in some software engineering benchmarks, like SWE-bench Pro (67.7 vs. Fable 5’s 80.0), indicating that its claim to being second in overall performance is based on a selective subset of tasks. The open weights, set to be released next week, are intended for high-end hardware, not for individual self-hosting, and the licensing details remain unpublished, adding a layer of uncertainty.
For fifteen days the claim ran without a benchmark table. Today Alibaba published the table, the active-parameter count, and a weights timeline. The numbers are genuinely strong on the rows Alibaba chose — and twelve to fifteen points behind on the rows it didn’t.
▲ All performance figures: Alibaba’s own harnessThe claim shipped on a Sunday. The evidence shipped two weeks later. In between, the claim did its work.
“Second only to Fable 5” is true on the rows Alibaba chose and false on the rows it didn’t. Both halves below are from the same release.
“Qwen3.8 is going open-weight” describes three things with very different deployment realities.
OpenAI- and DashScope-compatible — a base-URL change to A/B against your current backend.
A multi-node datacenter artifact. At 95B active, no single machine serves it. A flag planted, not a deployment option.
The checkpoint that fits real hardware. Whether the agentic gains survive distillation is the question that decides whether next week matters.
Three Chinese frontier releases in seventeen days, each measured against the same export-controlled model. The contest is real; it is not the same thing as your workload.
- The generation jump is real and consistent across a dozen agentic rows, with a stated mechanism: RL-environment scaling.
- More disclosure than Kimi K3 shipped — full table, active-parameter count, weights timeline.
- If 2.4T lands under a permissive licence, the ceiling of “open weight” moves permanently.
- The 27B sibling could become the best local agent model on hardware people already own.
- Every number is Alibaba’s harness. Independent testing already tempered Kimi K3’s launch claims substantially.
- The paying use case still belongs to Fable 5 — twelve to fifteen points on deep software engineering.
- “Next week” comes from a company that sat on a finished benchmark table for fifteen days.
- Until the licence text exists, “going open-weight” is a press strategy, not a property of the model.
and it says “second only” depends entirely on which row you read.
Implications of Alibaba's Model Performance Claims
The announcement of Qwen3.8-Max and its benchmark results mark a notable development in large-scale AI models, especially given Alibaba’s claim of being second only to GPT-5.6. This influences perceptions of China’s AI competitiveness and the potential for open-weight models to challenge dominant players like OpenAI. The detailed benchmark data provides transparency, but the selectivity of the metrics and the upcoming open-weight release will determine its real-world applicability and influence on AI deployment strategies.
LLM Systems Engineering: Training and Building Large Language Models – Engineering AI Models Through Fine-Tuning, Continued Pretraining, and From-Scratch Development
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Recent Large-Scale Model Launches and Benchmark Trends
Over the past month, major AI companies have announced or previewed models exceeding 2 trillion parameters, including Alibaba’s Kimi K3 and others. The industry has seen a surge in multimodal and agentic AI capabilities, with benchmarks like Terminal-Bench and PaperBench becoming key indicators. Alibaba’s approach, initially stealthy, culminated in today’s detailed release, following a pattern of strategic announcements and selective disclosure. The open weights for Qwen3.8-27B, a smaller, more deployable version, are scheduled for next week, emphasizing Alibaba’s focus on practical, hardware-friendly models.
"Qwen3.8-Max demonstrates our advancements in multimodal AI and agentic capabilities, setting a new standard for large models."
— Alibaba spokesperson
Limitations of Current Benchmark Data and Open-Weight Details
While Alibaba has published comprehensive benchmark results, the model’s performance on untested tasks remains unknown. The upcoming open weights are designed for high-end hardware, not for general self-hosting, and the licensing terms are still unpublished. It is also unclear how well the agentic improvements will hold up in real-world, long-term applications, or whether the model’s performance across all benchmarks is representative of its overall capabilities.
Upcoming Open-Weight Release and Industry Benchmark Comparisons
Next week, Alibaba will release the open weights for Qwen3.8-27B, enabling wider testing and deployment. Industry observers will scrutinize its performance on real-world tasks and compare it against other leading models like GPT-5.6 and Claude. Further benchmark releases and independent evaluations are expected to clarify whether Alibaba’s claims translate into practical, scalable AI solutions or if the performance gaps in specific tasks limit its overall standing.
Key Questions
What does 'second only to Fable 5' mean in Alibaba’s claims?
It indicates that Alibaba considers Qwen3.8-Max to be the second-best performing large language model based on their benchmark results, primarily in certain tasks like Terminal-Bench and PaperBench, but this is selective and does not encompass all evaluation metrics.
Will the open weights be usable for individual researchers?
No. The open weights for Qwen3.8-27B are designed for high-memory, multi-node hardware, not for self-hosting on standard consumer machines. Licensing details are still unpublished, which may impact accessibility.
How does Qwen3.8-Max compare to other models like GPT-5.6?
In benchmark tests, Qwen3.8-Max scores highly on several tasks but trails GPT-5.6 at the top end. Its strengths lie in multimodal and agentic capabilities, but it does not outperform GPT-5.6 across all benchmarks.
What are the main limitations of the current data?
The benchmark results are selective, and performance on untested tasks remains unknown. The model’s practical deployment and licensing details are still evolving, which limits full assessment of its capabilities.
Source: ThorstenMeyerAI.com
