A Beijing-based artificial intelligence company, Moonshot AI, has unveiled a new model that it claims outperforms both OpenAI’s GPT-5 and Anthropic’s Claude Sonnet 4.5 — and it’s completely free to use. The company announced that its latest innovation, Kimi K2 Thinking, has achieved record-breaking scores in benchmarks that evaluate reasoning, coding ability, and agent-based intelligence.

A Leap for Open-Source AI

Moonshot AI has made the model accessible through its API and official website, allowing developers and enterprises to experiment with its full capabilities. Unlike most proprietary Western AI systems, Kimi K2 Thinking is open-source, signalling a bold move toward greater transparency and collaboration in the global AI ecosystem.

Also Read| Zoho’s Arattai App Falls from Top 100 After Swadeshi Hype Fades

According to the company, this advancement marks a pivotal shift — demonstrating that open-source innovation can rival or even surpass closed models developed by U.S. tech firms.

Benchmark Results Show Surprising Edge

The company reports that Kimi K2 Thinking scored 44.9 % on the “Humanity’s Last Exam,” a complex reasoning benchmark with over 2,500 test questions. This score, if verified, puts it ahead of both GPT-5 and Claude Sonnet 4.5, an unexpected result that has drawn global attention to China’s growing AI capabilities.

Also Read| Google Gemini AI Pro Free Plan Now Available to All Jio 5G Customers

Efficiency Meets Intelligence

What makes Kimi K2 Thinking particularly remarkable is its cost-efficient training process. The model was reportedly trained with a much smaller budget compared to leading Western AI systems, showing that strategic design and optimisation can compensate for massive computing power. This breakthrough could redefine how AI research evolves, focusing not just on scale, but on smarter, more resource-efficient architectures.

Democratizing Access to Advanced AI

For developers, researchers, and startups, Moonshot AI’s decision to make Kimi K2 Thinking free and publicly accessible represents a significant step toward democratising AI. It offers an opportunity for a broader global community to test, build, and innovate using high-performance tools without prohibitive costs. The move also suggests a shifting AI landscape where competition isn’t just about who spends more, but who innovates smarter and shares globally.