📊 Full opportunity report: Second Only To Fable 5: Qwen3.8-Max Finally Shows Its Numbers — And The Claim Gets Complicated on ThorstenMeyerAI.com — validation score, market gap, and execution plan.
TL;DR
Alibaba announced the broad availability of Qwen3.8-Max, revealing detailed benchmark results that position it as the second-best model after Fable 5. The model boasts 2.4 trillion parameters, with strong performance in multimodal and agentic tasks, though some limitations remain.
Alibaba officially released the full benchmark table for Qwen3.8-Max on August 3, confirming it has 2.4 trillion parameters and ranks as the second-highest model after Fable 5, based on their own tests. The company also announced that open weights for the model will be available next week, marking a significant step in AI model transparency and accessibility.
Following two weeks of speculation and partial previews, Alibaba published the detailed benchmark results for Qwen3.8-Max, revealing a model built on the Qwen3.5 architecture with a sparse mixture-of-experts design. The model features approximately 95 billion active parameters per query, operating within a 983,616-token context window, and supports multimodal input including text, images, and video.
The benchmark table shows Qwen3.8-Max achieving top scores on several tests, including Terminal-Bench 2.1 at 86.6, surpassing Claude Fable 5 and only behind GPT-5.6 Sol at 88.8. It also leads in PaperBench at 93.0 and performs well in multimodal and agentic tasks, with notable scores in OSWorld-Verified and Parametric CAD Bench. However, it trails significantly in deep software-engineering benchmarks like SWE-bench Pro and FrontierSWE, with gaps of 12-15 points compared to Fable 5.
The company emphasized that the 2.4 trillion parameters are sparsely activated, with the active parameters per query around 95 billion, indicating a model that combines large-scale capacity with efficiency. They also demonstrated the model’s ability to reproduce research results and outperform previous versions on long-horizon agent tasks, marking a major improvement over prior iterations.
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 Benchmark Results for AI Leadership
Alibaba’s detailed benchmark release confirms its position as a leading AI developer, with a model that rivals GPT-5 in several areas. The disclosure of the 2.4 trillion parameters and performance metrics signals a shift toward more transparent and competitive large-scale models. The availability of open weights next week will enable broader access and testing, potentially impacting AI deployment and innovation across industries. However, the model still shows limitations in software engineering tasks, highlighting ongoing challenges in specialized AI capabilities.
As an affiliate, we earn on qualifying purchases.
Background of Alibaba’s Model Development and Benchmarking Strategy
Over the past two weeks, Alibaba’s Qwen3.8-Max was shrouded in secrecy, initially revealed through a stealth preview and an anonymous model called “kaleb” on the Code Arena leaderboard. The model was confirmed during the World AI Conference in Shanghai on July 19, after a series of teasers and leaks. The company’s approach involved staged announcements, culminating in the full benchmark release on August 3, which included detailed performance data for the first time.
This follows a recent surge of large model launches, including Moonshot’s Kimi K3 with 2.8 trillion parameters, and reflects Alibaba’s strategic emphasis on multimodal and agentic capabilities. The model’s design leverages sparse mixture-of-experts architecture, aiming to balance size, efficiency, and performance, especially in long-horizon reasoning tasks.
"Qwen3.8-Max demonstrates our commitment to advancing multimodal and agentic AI capabilities, with open weights coming next week."
— Alibaba spokesperson
As an affiliate, we earn on qualifying purchases.
Remaining Questions About Model Licensing and Real-World Use
It is still unclear what the licensing terms for the open weights will be, as Alibaba has not yet published the license details. The actual performance of the 27B checkpoint in real-world deployment remains to be tested, especially regarding whether agentic gains are preserved after compression. Additionally, the long-term impact of the model’s limitations in software engineering benchmarks is uncertain.
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.
Next Steps: Open Weights Release and Broader Testing
Alibaba plans to release the open weights of Qwen3.8-Max next week, enabling researchers and developers to evaluate its capabilities firsthand. The upcoming release will likely trigger a wave of independent benchmarking and application development. Further updates on licensing, deployment, and performance in diverse tasks are expected in the coming months as the model is integrated into broader AI ecosystems.
AI model performance testing software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What are the key specifications of Qwen3.8-Max?
Qwen3.8-Max has 2.4 trillion total parameters with approximately 95 billion active parameters per query, built on a sparse mixture-of-experts architecture, supporting multimodal input and long context windows.
When will the open weights be available?
Alibaba announced that open weights for Qwen3.8-Max will be shipped next week, enabling broader access for research and deployment.
How does Qwen3.8-Max compare to other models like GPT-5 or Fable 5?
According to Alibaba’s benchmarks, Qwen3.8-Max ranks second after GPT-5.6 Sol on Terminal-Bench, and surpasses Fable 5 in several multimodal and agentic tasks, though it trails in deep software engineering benchmarks.
What are the main limitations of Qwen3.8-Max?
The model underperforms significantly on deep software-engineering benchmarks and its agentic capabilities may be affected by compression in the 27B checkpoint. Its licensing terms remain undisclosed.
Why is the benchmark release significant?
The detailed benchmark results validate Alibaba’s claims of high performance and transparency, potentially reshaping competitive dynamics among large AI models.
Source: ThorstenMeyerAI.com
