Ali has officially launched and made its latest model, Qwen3.8-Flash, available as open source. This model features an entirely new architecture. Despite having only 6 billion (6B) of its total hundred-billion parameters activated, it outperforms Claude Opus 4.6. Ali states that, thanks to a thorough revamp of both the architecture and training processes, the training cost of Qwen3.8-Flash has been drastically reduced by nearly 90% compared to Qwen3.7-Plus. The inference cost has also seen a significant decrease, with input costs as low as 1 yuan per million Tokens and output costs at 3 yuan, marking a reduction to just one-third of DeepSeek-V4-Flash's price. This achievement sets a new benchmark for global model cost-effectiveness.