Anthropic launched the cost-effective small model Haiku 4.5, programming on par with Sonnet 4, with expected revenue of 9 billion this year, aiming to nearly double next year

OpenAI's rival Anthropic is accelerating the enrichment of its model product library. On the 15th of this month, Eastern Time, the company launched a new version of its smallest model, Claude Haiku 4.5, which performs comparably to the cutting-edge medium model Sonnet 4 released a few months ago in key tasks such as programming, but at only one-third of the cost and more than twice the speed.

The release of the new model comes at a time when Anthropic is rapidly expanding. This AI startup, valued at $183 billion, has over 300,000 enterprise customers, with enterprise products accounting for about 80% of its revenue. The code generation tool Claude Code, launched earlier this year, is nearing an annualized revenue of $1 billion.

According to media reports released on Wednesday, Anthropic is expected to achieve an annualized revenue target of $9 billion by the end of this year, and the company has set even more aggressive targets for next year: under baseline conditions, annualized revenue is expected to exceed $20 billion, and in the best-case scenario, it could reach $26 billion, nearly doubling revenue compared to this year. A spokesperson for Anthropic later stated that the company's annualized revenue this month is approaching $7 billion.

The revenue growth trajectory of Anthropic highlights the strong demand for generative AI tools among enterprises and explains the enthusiasm of investors. In comparison, OpenAI crossed the $13 billion annualized revenue threshold in August this year and is expected to exceed $20 billion by the end of the year.

Performance Improvements and Cost Advantages

Claude Haiku 4.5 has demonstrated performance that surpasses its small designation in multiple benchmark tests.

According to test results disclosed by Anthropic, Haiku 4.5 scored 73% in the SWE-Bench validation test and 41% in the Terminal-Bench test. Although this is lower than the " world's strongest programming model" Sonnet 4.5 released by Anthropic at the end of last month, it is comparable to the performance of its predecessor Sonnet 4, OpenAI's GPT-5, and Google's Gemini 2.5. Haiku 4.5 also showed similar levels in benchmark tests for tool usage, computer operation, and visual reasoning.

Mike Krieger, Anthropic's Chief Product Officer, stated to the media on Wednesday that Haiku 4.5 "far exceeds its scale." In certain tasks, such as computer usage, this small model even outperformed the medium model Sonnet 4 that Anthropic launched nearly five months ago Krieger stated in a statement that Haiku will make a new deployment method possible for the first time. "Sonnet handles complex planning, while the sub-agent driven by Haiku executes quickly—this opens up a whole new category of AI applications in production environments. We provide users with a complete toolbox of agents, each model with a combination of intelligence, speed, and cost suitable for different workflows."

For paid users, the price of the Haiku model is typically about one-third that of the Sonnet model, while the Sonnet model is one-fifth the price of the Opus model. The API pricing for Haiku 4.5 is $1 and $5 per million input and output tokens, respectively. The model is now available to all free users, who can still choose to use Sonnet 4.5, but using Haiku 4.5 offers more capacity.

Collaboration and Application Scenarios

Sonnet 4.5 and Haiku 4.5 can work together, leveraging their respective strengths. Anthropic stated that Sonnet 4.5 can create multi-step plans to solve complex problems, while Haiku 4.5 can complete the sub-tasks within those plans.

Parallel models are particularly useful for businesses looking to use AI to handle long-term projects. Krieger gave an example: "You can have Haiku monitor financial data streams—because it is a smaller, cheaper, and faster model that can process in larger batches—and then pass its preliminary insights to Sonnet for deeper analysis."

The most direct application scenario may come from the software development tools sector, where Claude Code has been widely used, and latency is often a critical factor. Andrew Filev, CEO of Zencoder, stated in a statement provided by Anthropic that the new version of Haiku "opens up a whole new category of use cases."

Krieger revealed that, personally, although Haiku is not as intelligent as Sonnet, he has started using it by default on Claude, "especially for mobile applications, because the speed of getting answers is much faster."

Rapid Iteration and Market Competition

Haiku 4.5 is the latest product released by Anthropic in a series of recent intensive launches. This model was released just two weeks after Sonnet 4.5, and only two months after the launch of Opus 4.1. The previous version of Haiku was released in October 2024.

Krieger stated that Anthropic is developing another model, which may be an updated version of Opus, planned for release by the end of this year or early next year.

The fast-paced industry competition does not allow Anthropic much breathing room after a release. While training Sonnet 4.5, the company has already initiated the development of Haiku 4.5. "We are really going all out," Krieger said.

This Wednesday, media reports indicated that Anthropic's revenue milestone came after the San Francisco startup experienced a round of rapid financing and valuation growth. The company recently raised $13 billion in a Series F funding round led by ICONIQ, with a valuation reaching $183 billion, nearly doubling from a valuation of $61.5 billion in March Anthropic develops the Claude large language model series, competing with OpenAI's GPT series. The company emphasizes AI safety and building models for enterprise use cases, with its models fueling the growth of code generation startups like Cursor.

Anthropic's revenue growth reflects the continued strong demand for generative AI tools among enterprises. The company's expansion of enterprise products is driving growth, with over 300,000 enterprise and business customers contributing approximately 80% of revenue. The company sells model access through APIs, offering a variety of products including Claude Code.

Anthropic is also expanding sales to the government and markets outside the United States. In August of this year, the company announced it would provide the Claude model to the U.S. government for a symbolic price of $1. The company plans to open its first office in Bangalore, India in 2026—India being its second-largest market after the U.S.—and aims to double its international workforce this year, expanding its AI application team fourfold to meet the growing demand