First Release! Qwen Office Launches Qwen3.8-Flash, Ushering in an Era of Abundant Agent Capacity

Task generation speed has nearly doubled, while average token consumption has dropped by 75%.

QwenWork Product

Today, Tongyi Qianwen Office has officially become the first to deploy the newly released Qwen3.8-Flash model and simultaneously launched its standard service mode.

From now on, all users can experience Qwen3.8-Flash through the brand-new standard mode.

Leveraging the latest model technology, users can complete tasks at a lower point cost and with significantly improved token throughput efficiency.

Looking ahead, Tongyi Qianwen Office will offer models in two distinct tiers: standard and advanced. Approximately 95% of everyday workloads can be handled by the standard mode, with only about 5% of highly complex tasks requiring the advanced mode.

The significant improvement in user experience is primarily attributed to model upgrades and synergistic optimization between agents.

Powered by a new architecture, Qwen3.8-Flash achieves performance surpassing Claude Opus 4.6, backed by hundreds of billions of total parameters.

Meanwhile, the Tongyi Qianwen large model team has collaborated with the Office team to launch a specialized office version of Qwen3.8-Flash. This version undergoes dedicated training and fine-tuning for specific scenarios such as multi-step planning, tool selection, and context compression. Through inference optimization and a custom-built Harness architecture, throughput performance is further enhanced.

Real-world office environment testing shows that under Qwen Office's standard mode, single-task generation speed has nearly doubled, while token consumption has decreased by an average of 75%.

In practical AI applications, high performance often comes with high costs and long latency, whereas pursuing low cost may compromise intelligence levels. The deep collaboration between agents and models is gradually breaking down the "impossible triangle" among performance, cost, and speed.

As model intelligence density continues to rise, combined with bidirectional optimization between Qwen Office and the underlying models, agents are poised to leave token anxiety behind and enter a new era of abundant and stable supply.

We dedicated to serving clients with professional DingTalk solutions. If you'd like to learn more about DingTalk platform applications, feel free to contact our online customer service or email at This email address is being protected from spambots. You need JavaScript enabled to view it.. With a skilled development and operations team and extensive market experience, we’re ready to deliver expert DingTalk services and solutions tailored to your needs!