Qwen 3 and 3.5: Alibaba enters the fight for the top

Alibaba didn't usually appear in the front lines of AI discussions. That changed in 2025. With the launch of Qwen 3 still in 2025 and Qwen 3.5 in February 2026, the Chinese company positioned its models among the most capable in the world — and opened access to them.

The history of the Qwen family is a story of intelligent scaling, multilingualism, and advanced reasoning.

Qwen 3: The Model That Leads in Reasoning

Qwen 3 235B is the largest and most capable in the family. With 235 billion total parameters, it stands out in scientific and mathematical reasoning benchmarks:

• GPQA Diamond (doctoral-level scientific reasoning): 77.2% — best open source in this category
• AIME '24 (mathematics olympiad): 85.7%
• Multilingualism: native support for dozens of languages, including Portuguese

Qwen 3's advantage isn't just size. It's the ability to follow complex instructions in multiple languages consistently — something many larger models still struggle with.

Qwen 3.5: Scaling with MoE

Launched in February 2026, Qwen 3.5 adopts Mixture of Experts architecture with 397 billion total parameters, activating only 17 billion per forward pass. Execution cost is manageable; capacity is frontier-level.

The license is Apache 2.0 — fully permissive for commercial use, fine-tuning, and redistribution, without royalties or restrictions. An important differentiator against other models in this category.

Its technical highlight is the context window: over 1 million tokens. This puts Qwen 3.5 in direct competition with Llama 4 Scout and positions it among the largest context models available in open source.

Compared directly with Llama 4 Maverick and first-rate proprietary models, Qwen 3.5 competes on equal footing in reasoning and coding tasks.

Why Does Alibaba Invest Heavily in Open Source LLMs?

The answer is strategic. Alibaba Cloud is one of the largest cloud platforms in Asia. High-quality open source models attract developers, who bring workloads to the company's cloud.

Additionally, Chinese regulation favors models under domestic control — which creates a massive protected market for Qwen products.

Practical Applications

Qwen 3 has been adopted at scale in:

• Analysis of legal and financial documents (high accuracy in instruction following)
• Multilingual corporate assistants (native support for Asian and European languages)
• Autonomous research agents (chained reasoning in scientific texts)

For teams working with data in multiple languages or needing advanced scientific reasoning, Qwen 3 is today one of the most solid choices in open source.

Conclusion

Alibaba transformed the Qwen family into a genuine global-level competitor. With leadership in GPQA Diamond and 1M+ token context, Qwen 3 and 3.5 models aren't just open source alternatives — they're benchmarks in their category.

The open source AI field has never been so competitive. And Alibaba makes sure to make this clear.

Sources:

• Hugging Face — Qwen3.5-397B-A17B
• MarkTechPost — Qwen 3.5 Release
• NVIDIA NIM — Qwen3.5-397B-A17B
• BenchLM — Best Open Source LLM 2026

Published on Hive.blog | #ArtificialInteligence #llm