Alibaba’s AI division, Tongyi, has released its web AI agent, WebSailor, as an open-source project.

The agent achieved the highest score on the BrowseComp benchmark, surpassing models such as DeepSeek R1 and Grok-3. BrowseComp is designed to test the reasoning and retrieval capabilities of web AI agents in complex scenarios.

The project is available on GitHub, allowing developers and researchers to explore and build upon the technology.

This move reflects a growing trend of tech companies open-sourcing AI tools to promote collaboration and innovation in the field.

🔗 Source: 36Kr

🧠 Food for thought

1️⃣ Strategic open-sourcing reflects shifting AI business models beyond proprietary advantage

Alibaba’s WebSailor release follows a clear industry pattern where companies increasingly view open-sourcing as strategically beneficial rather than a competitive disadvantage.

A comprehensive 2023 survey found that 76% of technology leaders expect to increase their use of open-source AI technologies in the coming years, reflecting the growing recognition of its business value 1.

This trend is particularly pronounced in the technology sector, where McKinsey research shows 72% of companies are already utilizing open-source AI models 2.

The shift occurs as companies recognize that open-sourcing creates multiplier effects: transparency builds trust, community involvement improves models, and broader adoption expands potential applications beyond what a single organization might develop.

For Chinese tech giants like Alibaba, open-sourcing also serves as a counterbalance to Western AI dominance, potentially accelerating global adoption of their technological frameworks and standards.

2️⃣ Benchmark competitions reflect the intensifying global AI race

Alibaba’s emphasis on WebSailor outperforming models like DeepSeek R1 and Grok-3 on the BrowseComp benchmark highlights how performance metrics have become crucial competitive battlegrounds in AI development.

Comprehensive benchmarking data shows that model performance correlates strongly with estimated training compute resources, with significant performance jumps occurring past certain computational thresholds 3.

These benchmarks have become proxy measures for technological advancement, with US models currently outperforming non-US models on key metrics, though the performance gap is narrowing with each new release 3.

Recent Alibaba developments