Alibaba Releases Open-Weights AI Model Qwen3.8-Max with Top Scores in Coding Benchmarks

In internal benchmark comparisons, Qwen3.8-Max performs significantly better than its predecessor Qwen3.7-Max in several tests. On the Terminal Bench 2.1 platform, the new model achieves a score of 86.6, ahead of its predecessor at 74.5. On SWE-bench Pro, Qwen3.8-Max reaches 67.7 points, compared to 60.6 for Qwen3.7-Max. Significant gains over the previous version are also evident on DeepSWE 1.1 (56.6 vs. 21.6), FrontierSWE (73.5 vs. 40.7), and PaperBench (93.0 vs. 64.8).

Benchmark Results at a Glance

In direct comparison with the current leading commercial models, Qwen3.8-Max delivers mixed results. On Terminal Bench 2.1, it scores 86.6 points, ahead of Anthropic's Claude Fable 5 (84.6) and behind OpenAI's GPT-5.6 Sol (88.8). On SWE-bench Pro, GPT-5.6 Sol (80.0) and Claude Fable 5 (69.2) lead the new Alibaba model (67.7). On the FrontierSWE test, Qwen3.8-Max takes second place with 73.5 points behind GPT-5.6 Sol (88.8). On GPQA Diamond, a test for logical reasoning, Qwen3.8-Max achieves 92.6 points, tying with Claude Fable 5 (92.6) and trailing GPT-5.6 Sol (94.1).

Benchmarks are standardized tests used to measure the performance of AI models on clearly defined tasks such as code generation or logical reasoning. They enable comparisons between systems from different providers, but can only to a limited extent reflect real-world relevance, as practical tasks deviate from the test environment. Industry observers nevertheless view the release as a signal that the gap between Chinese and US AI developments has narrowed.

Open Weights as a Strategic Approach

With its open-weights strategy, Alibaba is pursuing a different approach than its direct US competitors. While OpenAI relies on closed systems with GPT-5.6 Sol, whose internal weights are not published, Alibaba grants external developers access to the model parameters. This facilitates independent customization, local deployment, and independent security audits. Industry experts see this as a competitive advantage in markets with requirements for data sovereignty and data control, such as Europe.

The release is part of a series of Chinese model launches. Alongside Qwen3.8-Max, the model Kimi K3 is also positioning itself as a relevant competitor in the open-weights segment. Both systems are developed by Chinese providers that are increasingly relying on open releases to reach international developer communities and help shape technological standards.

Availability and Independent Validation

Alibaba has announced that the model will be made available via its cloud platform as well as via the platform Hugging Face, where developers can download AI models and collaboratively develop them further. As a result, Qwen3.8-Max is available for testing and applications outside the Alibaba infrastructure immediately after release. The company has not provided detailed information on specific licensing terms and permitted commercial uses in its announcement to date.

The benchmark scores come from four independent reports that consistently rank the Qwen3.8-Max model with the stated point values. Validation by independent third parties outside of Alibaba's tests is still pending for the results cited in the announcement. Industry observers emphasize that independent replications of the benchmark results must be awaited before the performance of Qwen3.8-Max can be conclusively assessed.

With Qwen3.8-Max, Alibaba is intensifying competition in the segment of AI models for software development. Should the benchmark results be confirmed in independent tests, the model would represent technological competition to established systems from the US and further diversify the global AI landscape.

The release is being received with interest in the European developer community, as open-weights models are particularly relevant under the requirements of the General Data Protection Regulation (GDPR) and amid concerns about data transfer to third countries. With Qwen3.8-Max, developers in Europe now have access to another powerful open-weights model suitable for local deployment in European data centers.