Ali released and synchronized the latest open source model Qwen3.8-Flash. The model uses a new next-generation architecture. Activating only 6 billion of 100 billion total parameters can obtain cutting-edge performance that surpasses Claude Opus 4.6, setting a new global benchmark for model efficiency. Thanks to innovations in architecture and training, Qwen 3.8-Flash training costs plummeted by nearly 90% compared to Qwen 3.7-Plus, and inference costs were also drastically reduced. The input cost was only 1 yuan per million tokens, and the output was 3 yuan. The price was as low as 1/3 of DeepSEEK-v4-Flash. Qwen3.8-Flash will launch “Qianwen Office” for the first time tonight. Developers and enterprises can also obtain the new model API service through the Qianwen AI platform.

Ali released and synchronized the latest open source model Qwen3.8-Flash. The model uses a new next-generation architecture. Activating only 6 billion of 100 billion total parameters can obtain cutting-edge performance that surpasses Claude Opus 4.6, setting a new global benchmark for model efficiency. Thanks to innovations in architecture and training, Qwen 3.8-Flash training costs plummeted by nearly 90% compared to Qwen 3.7-Plus, and inference costs were also drastically reduced. The input cost was only 1 yuan per million tokens, and the output was 3 yuan. The price was as low as 1/3 of DeepSEEK-v4-Flash. Qwen3.8-Flash will launch “Qianwen Office” for the first time tonight. Developers and enterprises can also obtain the new model API service through the Qianwen AI platform.