Qwen3 Undertakes Chatbot Arena Top 3; Compact Qwen3-30B Series Launches for Efficient AI Development

This article introduces Alibaba's Qwen3 models achieving a top-3 ranking on Chatbot Arena and the launch of a compact Qwen3-30B series for efficient AI development.

This week, Alibaba’s Qwen3-235B-A22B non-thinking mode (Qwen3-235B-A22B-Instruct-2507), secured third place on the competitive Chatbot Arena leaderboard just days after its launch; Meanwhile, Alibaba released its ungraded Qwen3-30B series with three streamlined models optimized for cost efficiency and development flexibility, empowering developers worldwide to harness cutting-edge AI without comprising performance.

Qwen3-235B Undertakes Top 3 on Global LLM Leaderboard

Qwen3-235B-A22B-Instruct-2507, Alibaba’s latest open-source large language model featuring 235 billion parameters and significant improvements in general knowledge, has secured third place on Chatbot Arena—tying with GPT-4.5 and Grok-4. Additionally, Qwen3 also ties for the first in Coding, Hard Prompts, Math, Instruction Following and Longer Query on Chatbot Arena as the top open model.

1_jpeg
Qwen3 Undertakes Chatbot Arena Top 3

Chatbot Arena, a widely respected benchmark platform for LLMs, evaluates models through anonymous comparisons in a crowdsourced and randomized format. This achievement further underscores the strong performance and growing competitiveness of Alibaba’s open-source language models in the global AI landscape.

2_jpeg
Qwen3 ties for the first in Coding, Hard Prompts, Math, Instruction Following and Longer Query as the top open model

With 256K long-context understanding capabilities, Qwen3-235B-A22B-Instruct-2507 exhibits significant improvements in general capabilities, including instruction following, logical reasoning, text comprehension, mathematics, science, coding and tool usage. It also showcases remarkable alignment with user preferences in subjective and open-ended tasks, enabling more helpful responses and higher-quality text generation.

New Qwen3-30B Series Delivers Powerful Performance in Small Sizes

In addition to open sourcing top-notched powerful large language models, Alibaba has been releasing upgraded lightweight yet powerful smaller-size models, which are easier to be deployed and run more cost efficiently for global developers. This week, the company unveiled a series of Qwen3 models with 30 billion parameters, including a thinking model (Qwen3-30B-A3B-Thinking-2507), a non-thinking model (Qwen3-30B-A3B-Instruct-2507) and a streamlined coding model (Qwen3-Coder-30B-A3B-Instruct).

Qwen3-30B-A3B-Thinking-2507: With enhanced capabilities in 256K long-context understanding, the model showcases markedly better alignment with user preferences, enabling more helpful responses and higher-quality text generation.
Qwen3-30B-A3B-Instruct-2507: With improved quality and depth of reasoning capabilities, the model exhibits remarkable general capabilities, such as instruction following, tool usage, text generation, and alignment with human preferences.
Qwen3-Coder-30B-A3B-Instruct: The model is featured with significant performance among open models on agentic coding, agentic browser-use, and other foundational coding tasks.

This article was originally published on Alizila written by Crystal Liu.

0 1 0

Share on

Community

Qwen3 Undertakes Chatbot Arena Top 3; Compact Qwen3-30B Series Launches for Efficient AI Development

Qwen3-235B Undertakes Top 3 on Global LLM Leaderboard

New Qwen3-30B Series Delivers Powerful Performance in Small Sizes

Read previous post:

Read next post:

Alibaba Cloud Community

You may also like

Comments

Alibaba Cloud Community

Related Products

Tongyi Qianwen (Qwen)

Alibaba Cloud for Generative AI

AI Acceleration Solution

Platform For AI