Replit is no longer working with Anthropic, CEO Amjad Masad confirmed this week. The coding platform announced a partnership with OpenAI back in August, but there was never any public signal that it had ended its relationship with Claude’s maker.

Masad said the decision came down to multiple factors, including cost. That’s a telling admission. Replit isn’t some hobbyist developer tool anymore. It’s a platform where millions of people write code, and every API call adds up fast. When you’re routing that much traffic through a third-party AI provider, pricing matters as much as performance.

The switch also highlights something bigger: enterprise AI partnerships aren’t sticky the way traditional enterprise software deals used to be. Companies can swap out model providers without rebuilding their entire stack. That’s good for customers, but it puts constant pressure on AI labs to compete on both capability and price.

Replit’s move is part of a broader trend. We’ve seen companies hedge their bets by supporting multiple models, and we’ve seen them consolidate around a single provider when costs get out of hand. What we’re starting to see now is companies making hard choices about which lab they want to bet on long-term.

It’s worth noting that Anthropic has been positioning itself as the more careful, more “enterprise-friendly” AI lab. But careful doesn’t always mean cheap, and it doesn’t always mean faster. OpenAI has been aggressive on both pricing and shipping new features. If you’re Replit and you’re trying to build features that feel magical to developers, you need a partner that can move fast and won’t blow up your burn rate.

Masad has been vocal about his vision for what he calls “self-driving companies,” where AI agents handle routine tasks and humans focus on higher-level strategy. That vision requires a lot of compute, a lot of inference, and a model provider that can scale with you. It also requires a partner that’s investing heavily in the kind of coding-specific capabilities that Replit’s users care about.

The timing is interesting, too. This comes just as the AI industry is starting to see real revenue concentration. A handful of big customers account for a huge chunk of API usage at the major labs. Losing one of those customers, especially one as high-profile as Replit, is more than a symbolic hit. It’s a signal to other platforms that switching costs are low and that labs need to earn their business every quarter.

Anthropic hasn’t commented publicly on the split, and it’s not clear whether Replit gave them a chance to match OpenAI’s offer. But the fact that cost was part of the equation suggests this wasn’t just about model quality. It was about economics.

For developers using Replit, this probably won’t change much in the short term. The platform will keep working, and most users won’t notice which model is running under the hood. But for the AI labs themselves, this is a reminder that enterprise customers are more willing to switch than anyone expected a year ago. The model provider that wins isn’t necessarily the one with the best benchmarks. It’s the one that can deliver the right mix of performance, price, and partnership.