Relace Apply 3

Relace Apply 3 is a code-editing language model from Relace, positioned within the company's Fast Apply series of specialized code-patching models and described as succeeding earlier Relace Apply generations. Its purpose is narrow and pragmatic: take diffs or edit snippets produced by upstream code models such as GPT-4o or Claude and merge those suggestions directly into existing source files, rather than regenerating whole files. This makes it well suited to automated refactoring pipelines, large-codebase patching, and code-agent workflows where a faster, dedicated applier steps in after a general-purpose model has drafted the change, helping preserve formatting, structure, and intent across the edited regions.

The model's design favors long-context engineering work, supporting both input and output as text and operating inside a very large working window that lets it ingest substantial codebases in a single merge pass. Aggregator listings describe it as a high-throughput specialist aimed at streaming and applying changes at speed while maintaining precision on the local edit. Because the weights are not publicly released, it is only reachable through hosted APIs, so teams integrating it typically pair it with a general-purpose coding model upstream and route the resulting edit through Relace Apply 3 to handle the precise file-level application step.