Alibaba's Qwen3.8 tops Artificial Analysis leaderboard for Agentic capability

Alibaba's Qwen3.8 has ranked first globally in Agentic capability on the Artificial Analysis leaderboard, according to a report from Chinese tech media QbitAI. The result places the open-source model ahead of leading domestic and international large language models in agent-related performance.

Artificial Analysis is a widely used third-party platform for cross-comparing large models, covering dimensions such as language, reasoning, coding, math and agentic capability. The Agentic score evaluates how well a model handles tool calls, multi-step planning and autonomous execution in real-world tasks, a key measure of whether a model can actually get work done.

As the newest open-source model from Alibaba's Tongyi Lab, Qwen3.8 builds on the influence of the Qwen family in the open-source ecosystem. Previous Qwen models have topped several public leaderboards, and this Agentic result strengthens their standing in agent application scenarios.

Agentic capability is becoming the next major battleground in the large model race. As AI agents move into production, a model's ability to reliably call tools and complete multi-step tasks directly determines the quality of downstream applications. Topping this dimension suggests Chinese open-source models can now compete head-on with the best international models at the agent infrastructure level.

For developers, the ranking offers a reference point for model selection. Models with high Agentic scores tend to mean lower debugging costs and more consistent task completion when building agent applications, and Qwen3.8's open-source nature lowers the barrier to adoption.

It is worth noting that leaderboard results reflect performance on specific evaluation dimensions, and real-world experience can vary across scenarios. Whether Qwen3.8 maintains its lead in further evaluations and production use remains to be seen.

The next points to watch are whether Alibaba ships a stronger commercial version built on Qwen3.8, and whether the model sustains momentum in community downloads and derivative development. The competitive landscape of agentic benchmarks could also shift as more new models arrive.