Chinese AI Nova from DeepSeek costs over 100 times less to operate than competitors

The competition among major Chinese artificial intelligence companies has seen new developments this week. While Alibaba presented its most robust AI model to date, DeepSeek focused on a distinct strategy: significantly reducing the operational costs of its tools.

According to Reuters, recent announcements show that the technological race in China is not limited only to performance but also aims to make these technologies more accessible to both developers and businesses.

Alibaba presents its most ambitious model

Qwen3.8-Max represents Alibaba's new initiative to compete among the most advanced AI systems available on the market. This launch generated significant buzz and boosted the company's shares by 7% on the Hong Kong Stock Exchange.

This technology integrates 2.4 trillion parameters, configured during training to allow for pattern recognition, response generation, and execution of various tasks. Although this volume is not the sole determinant of a system's quality, it remains a benchmark parameter for measuring the scale of the latest models.

Immediately after the launch, Qwen3.8-Max reached the top position among Chinese models in the Arena.AI text ranking. In the area of image analysis and other visual content, it achieved second place globally.

Alibaba highlighted several key features of this model, including processing of text, images, and videos; the ability to analyze up to 1 million tokens simultaneously; an architecture called 'mixture-of-experts,' which activates only a fraction of the model in each activity; and the concurrent use of only 95 billion parameters, resulting in lower cost and faster response time.

The company also announced that the new model will be available next week and that it completed a software engineering project in just 16 days.

Betting on open models

A common feature of both Qwen3.8-Max and DeepSeek's V4-Flash is the use of open-weight models. This allows developers to download these parameters and fine-tune them for various applications.

Chinese AI companies have identified a relevant market niche, given that many corporate routines do not necessarily require the industry's cutting-edge model.

Lian Jye Su, Chief Analyst at Omdia, commented to Reuters that this move meets a crucial market need. He stated: 'They need models that are good enough, accessible, transparent, and available, and open-weight models help meet this demand.'

In contrast, companies like OpenAI, Anthropic, and Google continue to invest in closed-source models.

DeepSeek targets cost reduction

While Alibaba concentrated its efforts on increasing the size of the new model, DeepSeek opted to compete through price. Artificial Analysis classified V4-Flash as the most economical model to operate among the leading systems evaluated in performance tests.

The survey indicated a cost of US$ 0.14 per million input tokens (equivalent to about R$ 0.71) and US$ 0.28 per million output tokens (approximately R$ 1.42). On average per test, V4-Flash recorded a cost of approximately US$ 0.03 (about R$ 0.15), compared to US$ 0.86 (about R$ 4.36) for Kimi K3, US$ 1.86 (approximately R$ 9.43) for GPT-5.6 Sol, and US$ 3.15 (about R$ 15.97) for Claude Fable 5.

These announcements confirm that the rivalry among Chinese companies has transcended the mere dispute over larger models. Currently, efficiency and cost also define who will have a place in the next stage of artificial intelligence.

olhardigital.com.brJuly 30, 21:59

OpenAI reduces AI model prices by up to 80% to increase business accessibility

OpenAI announced on Thursday, the 30th, a significant decrease in the costs of its smaller and mid-sized artificial intelligence models. This measure aims to make the use of these technologies more viable for companies amid growing concerns about AI-related expenses.

Details of Price Cuts

The company, known for developing ChatGPT, implemented an 80% cut in the cost of the GPT 5.6 Luna model and reduced the price of the mid-tier Terra model by 20%. This change comes in an increasingly competitive market, where there are disputes with other companies and the emergence of lower-cost Chinese alternatives.

Impact on Processing Rates

The new pricing structure primarily affects the cost charged for data processing in OpenAI's smaller models. Specifically, the Luna model is now priced at 20 cents per million input tokens, compared to the previous value of US$ 1. The Terra model saw its price drop from US$ 2.50 to US$ 2 per million input tokens.

Regarding response generation, the Luna's cost was reduced from US$ 6 to US$ 1.20 per million tokens, while the Terra was adjusted from US$ 15 to US$ 12. The price of the company's main model, named Sol, remained unchanged.

Market and Competition Context

This reduction occurs during a period when many corporations are finding it challenging to estimate the financial impact of intensive AI use. The sector has shown a trend of transitioning from fixed subscription plans to usage-based billing systems, which can generate instability and increases in operational costs.

OpenAI attributes this possibility of price drops to efficiency advancements integrated into GPT 5.6, citing improvements in both code development and internal system performance optimization.

Competitive Pressure in the Sector

The pricing strategy also exerts pressure on competitors. Anthropic, which has a strong user base among developers and companies with its Claude line, maintains higher prices on the mid-tier model equivalent to Terra. Data shows that Claude Sonnet 4.6 costs US$ 3 per million input tokens and US$ 15 per million output tokens, values higher than those currently practiced by Terra.

Furthermore, the market dispute is intense with the growth of open-source models developed by Chinese competitors, such as GLM 5.2, linked to Z.ai, which is pointed out as an option capable of approaching the performance of major Western companies at a lower cost.

AI Sector Analysis

Analysts observe that the price decrease may encourage more organizations to adopt artificial intelligence tools. However, this movement may also exert financial pressure on companies in the sector before any potential public stock offerings.