Replit Drops Anthropic for OpenAI in Major AI Coding Shift
Replit CEO Amjad Masad cites performance and cost as the developer platform cuts ties with Claude in favor of GPT models.
In a dramatic shift for developer tooling, AI coding platform Replit has ended its working relationship with Anthropic, choosing instead to double down on its partnership with OpenAI. Replit CEO and co-founder Amjad Masad revealed onstage at TechCrunch Disrupt that the platform has transitioned its core underlying model infrastructure to OpenAI's Sol and Astra models, citing superior benchmarks, better pricing, and deeper partnership capabilities.
Key Details
The announcement marks a significant realignment in the generative coding landscape, where Anthropic's Claude models—particularly Sonnet and Opus—had long been considered the premier choice for developer assistants. Under its previous agreement with Anthropic, Replit relied on Claude to power its popular Replit Agent and cloud development environment, driving substantial revenue growth across its user base.
- OpenAI Migration: Replit has transitioned its production environment to run primarily on OpenAI's GPT-6.1 Sol and Astra models.
- Performance Benchmarks: Masad stated that internal and public Replit benchmarks consistently showed OpenAI models outperforming Anthropic's Claude Fable and Opus in building "vibe-coded" applications from scratch.
- Enterprise Growth: The shift comes as Replit's revenue mix rapidly evolves, with enterprise contracts growing from 2% of revenue last year to 15% today, and overall B2B revenue exceeding 50%.
- Conflicting Reports: While Masad publicly confirmed the end of the Anthropic partnership, sources close to Anthropic claim the relationship remains active, highlighting recent Replit documentation referencing Claude Fable 5.1.
What This Means
Replit's decision underscores the intense competition among foundation model providers to secure developer-facing platforms. As AI coding tools move from novelty code-completion helpers to fully autonomous software engineering agents, platform operators like Replit are continually re-evaluating model providers based on raw latency, cost per token, and complex reasoning performance.
By aligning exclusively with OpenAI, Replit is betting that models like GPT-6.1 Sol and Astra will maintain a decisive lead in end-to-end application generation. For Anthropic, losing a flagship developer partner like Replit represents a notable commercial blow, even as the lab continues to expand its enterprise presence with its own direct tools like Claude Code and Cowork.
Technical Breakdown
The transition highlights critical technical requirements for running autonomous coding agents at cloud scale:
- Token Economy and Latency: Building full-stack web applications from scratch involves processing tens of thousands of context tokens per prompt, making lower model inference costs and faster response times critical for maintaining margins.
- Agent Environment Sandboxing: Autonomous coding agents require a secure digital habitat equipped with robust sandboxing, error recovery loops, network isolation, and identity management to prevent unintended command execution.
- Vibe-Coding Benchmarks: Evaluation criteria have shifted from static code completion metrics to dynamic end-to-end execution benchmarks that measure how accurately an agent translates natural language requests into fully functional, deployed web software.
Industry Impact
This realignment is likely to trigger further volatility across the AI developer tool ecosystem. As model capabilities converge, developer platforms are gaining leverage over model providers, treating foundation models as interchangeable commodities based on real-time performance and pricing.
Furthermore, Replit's focus on enterprise and B2B workflows demonstrates that business users are rapidly replacing hobbyists as the primary revenue driver for AI coding tools. Software companies must now prove their agents can operate safely within strict corporate compliance and security boundaries.
Looking Ahead
As Replit expands its enterprise footprint, all eyes will be on whether OpenAI's Sol and Astra models can maintain their edge in complex software architecture generation. Meanwhile, the developer community will closely monitor how Anthropic responds to this high-profile departure and whether rival platforms follow suit in shifting their underlying model infrastructure.
Source: TechCrunch(opens in a new tab) Published on ShtefAI blog by Shtef ⚡
