In the world of artificial intelligence, two new models from Chinese companies have recently emerged with drastically lower price points, but their actual performance on critical tasks remains a concern. The first to appear was DeepSeek's V4 Flash, which launched at an attractive cost of only three cents per benchmark task. This makes it 105 times cheaper than Anthropic’s more expensive Claude Fable 5, which charges $3.15 for a similar level of work. However, the low cost comes with a caveat.
How cheap is too cheap for AI models?
DeepSeek's V4 Flash is not without flaws. Its performance is questionable, with a reported 37% accuracy rate and a 84% hallucination rate—meaning it generates incorrect or fabricated results most of the time. While the price may look appealing, users need to be cautious about the trade-offs they’re making for such a low cost. A balance between affordability and reliability is essential in evaluating these models.
On the other side, Alibaba is preparing to launch its Qwen3.8-Max model next week, which appears to offer more promise. With 2.4 trillion parameters, this will be one of the world's largest open AI models. It's designed to tackle complex, long-term tasks without human oversight. Alibaba claims the model demonstrated its capability by spending 16 days developing and refining a coding tool, proving its ability to work through extended projects independently.
Price wars and AI spending habits
In addition to its impressive technical specifications, Qwen3.8-Max is also priced competitively. Alibaba offers it at $2 per million input tokens and $6 per million output tokens. This cuts Anthropic's pricing in half on output and reduces it by 40% for input. Although Alibaba asserts that the model's accuracy is on par with Anthropic’s Fable 5 and OpenAI’s GPT5.6-Sol, no independent studies have yet verified these claims.
The competitive pricing landscape is not exclusive to Alibaba. OpenAI has also made waves by slashing its prices for the GPT-5.6 Luna and GPT-5.6 Terra models. Specifically, GPT-5.6 Luna is now 80% cheaper, while GPT-5.6 Terra's cost dropped by 20%. However, these developments may not be of much concern to many organizations, which appear to be struggling with managing their AI-related expenses. A recent study revealed that 29% of companies are spending more than a quarter of their cloud budgets on AI.
Market reactions and future challenges
The recent reduction in AI costs has played a role in lifting major stock indices. This boost was partly driven by optimism from Washington, D.C. regarding the situation in Iran, but it was also influenced by strong earnings reports from tech companies.
Palantir experienced a 30% stock surge following its latest financial results, which included record revenue growth. Amazon, too, saw its stock climb after reporting a significant increase in net sales, reaching $200.6 billion for the year. The company’s AWS business, according to Forbes contributor Peter Cohan, saw the fastest growth in 18 quarters with a 37.6% revenue increase. Still, uncertainty around AI spending and high capital expenditures remains a challenge for investors.
Elon Musk's SpaceX has been making headlines for AI-related spending. In its inaugural quarterly report, the company disclosed an AI-related capital expenditure of $15.8 billion, which is a sixfold increase compared to the previous year and well above the $13.2 billion estimate. Following this report, SpaceX’s stock dropped by 13%. Despite this, AI-related revenue surged by 247%.
Interestingly, the news about SpaceX’s future AI data centers brought a positive reaction to Nvidia. The company saw a 3% stock increase after SpaceX announced that it plans to rely solely on Nvidia chips for upcoming projects, signaling continued trust in Nvidia’s hardware for AI computing.