Event date · · Alibaba

Alibaba PAI released SearchQwen2.5-7B, an open-weight search agent model

Alibaba 阿里巴巴Chinese AIOpen weights

Model: Qwen2.5-7B-Instruct · availability, license and releases

FACT STATEMENT

Alibaba Cloud PAI released SearchQwen2.5-7B, a 7.62B-parameter search agent model based on Qwen2.5-7B-Instruct, on Hugging Face under Apache-2.0. It is trained with EasyDistill 2.0 and SynSearch-Data, and supports structured search/browse tool calls.

China context

Original name
SearchQwen2.5-7B
Outside China
Open weights · huggingface.co
Claims
Company-reported; not yet independently evaluated
For builders
Developers outside China can download the weights from Hugging Face and integrate the model into their own search pipelines, using the structured tool-call interface with any search/browse backend.
For investors
The release of an open-weight search agent model by Alibaba PAI may pressure commercial search API providers to differentiate on quality or price, as enterprises gain a self-hostable alternative.

Translated from Chinese. Quotes and facts link to the original sources.

What happened

Alibaba Cloud PAI released SearchQwen2.5-7B, a compact search agent model based on Qwen2.5-7B-Instruct, on Hugging Face under Apache-2.0. The model is trained with environment-aligned, solver-verified search trajectories generated by EasyDistill 2.0, and is designed for multi-hop search, browsing, and evidence integration. It supports structured search/browse tool calls and requires an external search/browse backend.

Technical significance

The model is a 7.62B-parameter transformer with a context length of 32,768 tokens, fine-tuned from Qwen/Qwen2.5-7B-Instruct. It uses a structured tool-call interface where the model returns a tool call, the tool is executed, and the response is appended until a final answer is produced. Reported LLM-judge accuracy: Multi-hop QA 55.90% (vs 45.23% for base), Deep Search 33.33% (vs 24.35%), Overall 44.61% (vs 33.33%).

Industry impact

Developers outside China can now use an open-weight search agent model under Apache-2.0, reducing the cost and complexity of building multi-hop search and browsing applications compared to proprietary APIs. Competitors offering search-augmented models must match this capability or risk losing developer mindshare.

Decision value

The model enables enterprises to deploy search agent capabilities on-premises or in private clouds without per-query API costs, potentially lowering total cost of ownership for search-intensive applications. The Apache-2.0 license permits commercial use and modification.

What to watch

Next signals to check: whether Alibaba PAI releases larger SearchQwen variants, whether the SynSearch-Data training set is made public, and whether third-party evaluations confirm the reported accuracy gains on multi-hop QA and deep search benchmarks.

Latest in Chinese AI

  1. DeepSeekReportedly: DeepSeek V1 author’s startup project is an open-source agent for local deployment
  2. DeepSeekReportedly: OpenDesign team's OpenMotion lets DeepSeek Harness create motion graphics
  3. XiaomiReportedly: Xiaomi adds MiMo UltraSpeed to Token Plan Pro and Max for a limited time
  4. MiniMaxMiniMax publishes cryptanalysis findings from its internal AI model
  5. DeepSeekReportedly: Simplex AI founder Tony Zhang open-sources Linger, a personal agent that can run locally

All China AI Events

AIGC Newsletter

China AI, with sources and context.

Analysis of Chinese AI models, companies and policy, and what you can use outside China.