--- title: "The AI competition has entered the \"cost-saving\" era, with various new models competing for cost efficiency" type: "News" locale: "en" url: "https://longbridge.com/en/news/292414738.md" description: "OpenAI, SpaceX AI, and Meta recently launched new models, shifting the competitive focus to cost efficiency. OpenAI's GPT-5.6 emphasizes reducing token usage; SpaceX AI's Grok 4.5 claims to double token efficiency; Meta plans to offer Muse Spark 1.1 at a highly attractive price. As companies tighten AI spending, the market is refocusing on the cost of model usage, and developers need to maximize customer value while ensuring profitability to recoup substantial infrastructure investments" datetime: "2026-07-12T23:52:58.000Z" locales: - [zh-CN](https://longbridge.com/zh-CN/news/292414738.md) - [en](https://longbridge.com/en/news/292414738.md) - [zh-HK](https://longbridge.com/zh-HK/news/292414738.md) --- # The AI competition has entered the "cost-saving" era, with various new models competing for cost efficiency OpenAI, SpaceXAI, and Meta launched new models one after another last week, with all three companies emphasizing advancements in reasoning capabilities and cost efficiency, shifting the competitive focus to who can accomplish more work at a lower cost. OpenAI stated that its most advanced GPT-5.6 model is designed to accomplish more tasks while significantly reducing token usage, making it more cost-effective for customers. After rebranding from xAI to SpaceXAI, the newly launched Grok 4.5 model claims to have token efficiency that is twice that of comparable models from other companies. Meta's CEO Mark Zuckerberg also mentioned that Meta will offer very attractive pricing for its latest Muse Spark 1.1 model. As enterprise customers strictly scrutinize AI spending, the market is refocusing on the cost of model usage. Earlier this year, many companies encouraged employees to use AI as much as possible, competing over who could use it more, a practice known as "tokenmaxxing." However, in recent months, after receiving high bills, many companies have begun to tighten usage restrictions, partly due to model developers like Anthropic switching to pay-as-you-go billing instead of simply charging a fixed subscription fee. Model developers now must find ways to maximize the value potential customers can obtain without reducing their own profits. Otherwise, they will struggle to recoup the hundreds of billions of dollars invested in AI infrastructure. With its advertising business as a backing, Meta is preparing to adopt a more aggressive expansion strategy. "The pricing from other AI labs is very high, and the profit margins are astonishing. We believe there is a real opportunity to provide advanced, top-tier AI capabilities at a more affordable price," Zuckerberg told Bloomberg. OpenAI may not have as much room for price adjustments as Meta, but it also believes it must remain competitive on costs. OpenAI CEO Sam Altman stated in an interview with CNBC last week, "Every company is now thinking about AI spending and what value these expenditures actually bring, which is exactly what we hope to achieve." In addition to launching new models with higher cost efficiency, OpenAI has also begun assisting companies in managing AI spending, launching credit usage analytics for ChatGPT Enterprise last month and updating spending control mechanisms. As companies increasingly prioritize AI spending, they are looking for other potential solutions, and the market is not lacking in alternative options. Among them, Chinese model developers represented by DeepSeek have launched lower-priced open-source models. Although these models still lag behind the advanced products developed by American companies, they are sufficient for many everyday tasks. Additionally, some companies and developers are beginning to turn to model routing, which can be seen as a task diversion mechanism within AI systems, assessing the type and complexity of tasks in real-time and automatically directing requests to the most suitable model, thus achieving the best balance between performance, processing speed, and token costs One of the companies providing such services, OpenRouter, completed over $100 million in financing in May this year, indicating that the market and investors are optimistic about its prospects. It is worth mentioning that when emphasizing cost efficiency, model developers may further pressure the leading company, Anthropic. According to data from the AI benchmarking service Artificial Analysis, Anthropic's Claude Opus and Claude Fable models are among the most expensive in the market when calculated by cost per task. Elon Musk directly targeted Anthropic in a post promoting Grok 4.5, stating, "Grok 4.5 is a Claude Opus-level model, but faster, more efficient in token usage, and at a lower cost." As OpenAI, SpaceXAI, and Meta take action, the market's evaluation criteria for models are also changing. The new round of competition will push AI into an era centered on cost efficiency. - OpenAI, Meta, SpaceX Compete for More Cost-Efficient AI Models (Source of the main image: Unsplash) ### Further Reading: - Powered by GPT-5.6 model! ChatGPT Work launched, completing workflows across applications - SpaceXAI releases new Grok 4.5 model, achieving double the token efficiency of comparable models - Muse Spark model update, promoting Meta Model API, Zuckerberg rarely tweets in promotion ### Related Stocks - [META.US](https://longbridge.com/en/quote/META.US.md) - [SPCX.US](https://longbridge.com/en/quote/SPCX.US.md) - [METW.US](https://longbridge.com/en/quote/METW.US.md) - [FBL.US](https://longbridge.com/en/quote/FBL.US.md) - [METD.US](https://longbridge.com/en/quote/METD.US.md) - [METU.US](https://longbridge.com/en/quote/METU.US.md) - [SPCH.US](https://longbridge.com/en/quote/SPCH.US.md) - [SPCF.US](https://longbridge.com/en/quote/SPCF.US.md) - [SPAL.US](https://longbridge.com/en/quote/SPAL.US.md) - [SNK.US](https://longbridge.com/en/quote/SNK.US.md) - [SSPC.US](https://longbridge.com/en/quote/SSPC.US.md) - [SPCU.US](https://longbridge.com/en/quote/SPCU.US.md) - [LOFF.US](https://longbridge.com/en/quote/LOFF.US.md) - [FBY.US](https://longbridge.com/en/quote/FBY.US.md) - [FBYY.US](https://longbridge.com/en/quote/FBYY.US.md) - [MAGX.US](https://longbridge.com/en/quote/MAGX.US.md) ## Related News & Research - [Meta Pulls AI Image Feature From Instagram Days After Launch](https://longbridge.com/en/news/292479961.md) - [Meta AI image detector fails to identify some of its own cropped AI images, Reuters analysis finds](https://longbridge.com/en/news/292339206.md) - [Meta discontinues AI image feature days after launch](https://longbridge.com/en/news/292362710.md) - [Mark Zuckerberg Posted on Elon Musk's X for the First Time Since 2023 to Promote Meta's Muse AI—MKBHD Explains How to Opt Out](https://longbridge.com/en/news/292275185.md) - [Meta u-turns on AI feature amid privacy backlash](https://longbridge.com/en/news/292388524.md)