Replit Ends Anthropic Partnership Over Cost and Performance
Replit has terminated its partnership with Anthropic, citing cost, performance, and strategic concerns. The AI coding platform will continue using multiple AI models, including Anthropic's, as it focuses on its multi-model strategy and enterprise growth.
Replit has ended its partnership with Anthropic. CEO Amjad Masad announced the termination at TechCrunch Disrupt on Thursday, citing cost, partnership dynamics, and model performance as the reasons. The company continues to use Anthropic models on its platform, but the formal partnership is over.
Masad stated that safe deployment of AI agents requires orchestration, error recovery, security, process isolation, networking capabilities, identity management, and authorization. Replit's own Agent product, which performs initial triage on incoming support tickets, is central to this vision. The multi-model flexibility strategy aims to prevent any single AI provider relationship from jeopardizing the company. Corporate contracts now account for approximately 15% of Replit's revenue this year, up sharply from about 2% last year.
Replit's multi-model strategy emerges
Replit's approach now centers on using multiple AI models rather than relying on a single provider. This shift is a strategic defense. Both Anthropic and OpenAI are seen as potential competitors that could launch tools targeting Replit's user base. The company's $9 billion valuation is seen as pricing in continued execution risk due to its concentration on third-party models. Yet, its revenue growth signals strong demand for AI-driven coding products.
Replit's financial growth and enterprise focus
Replit's business is shifting toward enterprise clients. More than half of its revenue is generated by the B2B segment. The growth in enterprise contracts has been rapid.
| Revenue Segment | Last Year | This Year (2026) |
|---|---|---|
| Enterprise Contracts | ~2% | 15% |
| B2B Segment | Not Specified | >50% |
This financial pivot shows the platform's evolution from a consumer-focused tool to a service for corporate development teams.
Internal benchmarking reveals model performance differences
Replit evaluates AI models using an internal performance benchmark. According to these tests, specific models excel at certain tasks. Masad said, "We publish a public benchmark as well, and consistently, Sol and Astra perform better on building vibe-coded applications from scratch." He also noted that "I think ChatGPT is better now," indicating shifting competitive performance. The internal benchmark is a key tool for selecting which models to use for different functions on the platform.
Accounts differ on the current state of the Anthropic relationship. One source familiar with the partnership said cooperation is still ongoing, and Replit has not made a public announcement about terminating the collaboration. Despite ending the formal partnership, recent public content from Replit shows it continues to use Anthropic models. The company has also run its self-driving company model internally since early 2026. Replit's continued development of its multi-model Agent product remains a core focus.
