Alibaba Cloud Introduces Major Price Cuts for AI Models

Alibaba Cloud Announces Significant Price Reductions for Qwen-VL Models

Alibaba Cloud has made headlines once again with a substantial price cut for its Qwen-VL series large models. This announcement marks the third round of price adjustments for the year, following reductions in May and September. The new pricing structure offers a remarkable decrease, with the Tongyi Qianwen series visual understanding models experiencing an overall reduction of over 80%.

The Qwen-VL-Plus model now costs only 0.0015 yuan per thousand tokens, a staggering 81% drop, while the higher-performance Qwen-VL-Max has been reduced to 0.003 yuan per thousand tokens, representing an 85% decrease. Under the new pricing model, users can process approximately 600 images at 720P or 1700 images at 480P for just 1 yuan.

Features of the Qwen-VL Series

The Qwen-VL series consists of multi-modal models that have gained popularity within the open-source community due to their robust visual reasoning capabilities. These models are designed to recognize images of various resolutions and aspect ratios, and they can comprehend extended videos exceeding 20 minutes. The versatility of Qwen-VL allows it to understand tasks performed by intelligent agents, such as mobile devices and robots, making it applicable across a wide range of visual recognition scenarios, including smartphones and automobiles.

Reasons Behind the Price Reduction

According to the Alibaba Cloud Bailian team, the recent price cuts are attributed to ongoing optimization of both the cloud infrastructure and model architecture. Additionally, the exponential growth in model usage has facilitated economies of scale. Continuous technological advancements and optimizations have also led to significant improvements in inference efficiency.

The introduction of an elastic AI computing power scheduling system, in tandem with the Bailian distributed inference acceleration engine, has effectively reduced model inference costs and accelerated inference times. Alibaba Cloud has noted that as the visual understanding capabilities of Qwen-VL improve, the model has emerged as one of the fastest-growing options on the Bailian platform.

New KV Cache Billing Model

To enhance cost efficiency for users utilizing the large model API, Alibaba Cloud Bailian has rolled out a new KV Cache billing model. This innovative model significantly lowers invocation costs by automatically caching context, thereby avoiding redundant computations. The KV Cache model is particularly advantageous for applications involving long texts, code completion, multi-turn dialogues, and specific text summarization.

Implications for AI Accessibility

With these price reductions, Alibaba Cloud is not only making AI technology more accessible but is also opening up new opportunities for developers and enterprises. By continuously enhancing performance and reducing usage costs, Alibaba Cloud is driving the widespread adoption and application of AI technology, which provides robust technical support for the digital transformation across various industries.

Key Points

1. Alibaba Cloud has reduced prices for its Qwen-VL series models by over 80%.
2. Users can now process 600 images for just 1 yuan.
3. The price reduction is attributed to infrastructure optimizations and increased model usage.
4. A new KV Cache billing model has been introduced to further reduce costs for users.
5. These changes promote greater accessibility to AI technologies for developers and businesses.