A new open source AI model from Chinese startup Moonshot AI has matched or exceeded the performance of leading proprietary systems, including OpenAI’s GPT-5 and Anthropic’s Claude Sonnet 4.5, according to multiple benchmarks.
Carl Franzen reports for VentureBeat that the Kimi K2 Thinking model achieved 44.9 percent on Humanity’s Last Exam, 60.2 percent on BrowseComp, and 71.3 percent on SWE-Bench Verified. These scores surpass those of GPT-5 and other paid models on key reasoning and coding tasks.
The model uses a Mixture-of-Experts architecture with one trillion parameters, of which 32 billion activate during each inference. Moonshot AI released it under a Modified MIT License that allows free commercial use with one restriction: applications serving over 100 million monthly users or generating over $20 million monthly revenue must display “Kimi K2” on their interface.
The release follows a similar model from Chinese competitor MiniMax AI just weeks earlier. Both open source systems now rival proprietary alternatives while costing significantly less to operate. Moonshot prices K2 Thinking at $0.60 per million input tokens, compared to GPT-5’s $1.25.
The development adds pressure on U.S. AI companies to justify their massive infrastructure investments. OpenAI recently faced scrutiny after its CFO mentioned potential government support for the company’s $1.4 trillion in compute commitments, though the company later clarified it was not seeking federal backing.
Enterprises can now access frontier-level AI performance through free open source models rather than expensive proprietary services.
