OpenAI has upgraded its free user tier to default to the GPT-5.6 Luna model, enabling unlimited text chats for roughly one billion ChatGPT users worldwide. This move comes just one day after DeepSeek announced plans to significantly raise its API service pricing. The contrasting strategies of offering free access versus increasing prices highlight how the rise of open-source models is reshaping the global AI pricing structure.

This decision to go free follows OpenAI's price cut for GPT-5.6 Luna just a week ago on July 30, when it slashed output costs by 80% from $6 per million tokens to $1.20, and input costs from $1 to $0.20. On the same day, DeepSeek launched the official version of its V4 Flash API. Despite the price reduction, developer migration continued. Data from the third-party model aggregation platform OpenRouter shows that from August 3 to 7, DeepSeek V4 Flash (including both old and new versions) processed 11.31 trillion tokens, while GPT-5.6 Luna processed 3.54 trillion tokens—a nearly threefold difference. It is important to note that the OpenRouter data reflects only calls distributed through that platform, excluding direct API calls from OpenAI. However, this gap still indicates that the price cut failed to curb the developer shift toward open-source models. Data from the open-source AI coding platform OpenCode further reveals that on August 1 alone, DeepSeek V4 Flash handled 8 trillion tokens on its platform.

The logic behind developer choices is straightforward. A comparison report from Huatai Securities shows that the mixed price for DeepSeek V4 Flash is about $0.06 per million tokens, roughly 65% lower than OpenAI’s GPT-5.6 Luna, and the cost per task is about 57% lower. With comparable performance, the cost difference is decisive.

With price cuts failing to stem the tide, OpenAI has shifted to a more aggressive free strategy. Luna, an entry-level, lightweight model in the GPT-5.6 family, emphasizes speed and cost efficiency. Compared to the previous default model, GPT-5.5 Instant, Luna reduces factual error rates by 62%. Free users can also access more advanced reasoning capabilities through a new "Think" button. On the paid side, the GPT-5.6 Sol model handles revenue generation: Plus and Pro users can freely adjust a thinking slider, unifying the previously separate modes of Instant quick answers and Thinking deep reasoning. This dual approach uses Luna as a free gateway to attract a massive user base and data, while Sol maintains the paid revenue base, essentially reviving an internet-era strategy of free traffic.

This strategy shift signals that the rise of domestic open-source models poses a credible threat to the pricing power of Silicon Valley tech giants. Although DeepSeek V4 Flash quickly topped the third-party global usage charts, the widespread adoption of AI agents has led to a single task consuming up to a hundred times more computing power than a standard conversation. Peak-time computing resource contention has directly caused service instability, turning the low-price strategy into a scenario where higher usage leads to greater losses. According to a report from Kaiyuan Securities, in the first quarter of 2026, Zhipu AI raised its API prices by a cumulative 83% while seeing a 400% increase in usage. Earlier this year, major cloud providers like Alibaba Cloud, Tencent Cloud, and Baidu AI Cloud also collectively raised prices for AI-related computing power. DeepSeek's shift in direction shows a strategic pivot for large-model companies, moving from using low prices to capture market share toward a value-based pricing model, where token pricing logic is returning to a focus on value.

Amid this pricing power reshuffle triggered by the impact of open-source models, the industry's debate over the values of open versus closed systems is also heating up. Hugging Face CEO Clem posted that Google could have dominated the AI industry by open-sourcing frontier technologies like Gemini, but instead restricted them behind APIs for billions in revenue. In response to Clem's post, Turing Award winner Yann LeCun commented that a key reason he chose to join Meta in 2013 rather than Google was the company's open culture, arguing that AI cannot advance in a closed environment. LeCun noted that his eventual departure from Meta was also due to a change in the company's core culture. The rise of domestic open-source models is rewriting the global rules of AI competition, and the power structure of the industry is undergoing a fundamental transformation.