Morph

Morph: Inference Built for Coding Agents

Morph builds specialized models and APIs that make coding agents faster and cheaper, and its flagship is Fast Apply: a model trained specifically to merge AI-generated edits into existing source files. Instead of asking a frontier LLM to rewrite an entire file for every change, an agent sends Morph the original file plus a compact edit snippet with "existing code" markers, and Fast Apply returns the fully merged file at 10,500+ tokens per second with about 98% accuracy.

It is aimed at teams building coding agents, IDE assistants, and code-generation products. You call the API directly (or through OpenRouter), or drop in Morph's plugins for Claude Code and OpenCode. Around Fast Apply, Morph has grown a suite for agent infrastructure: WarpGrep for agentic codebase search, FlashCompact for context compaction at 25,000+ tokens per second, and Reflexes for agent behavioral observability.

Key Features

• Fast Apply model merging code edits at 10,500+ tokens per second with roughly 98% accuracy
• 50-60% token savings and 90%+ latency reduction versus full-file rewrites — a 1,000-line file applies in about 1.3 seconds
• Two model sizes (morph-v3-fast and morph-v3-large) with published per-token pricing from $0.80/M input
• WarpGrep code-search subagent and FlashCompact context compaction for agent pipelines
• Integrations and plugins for Claude Code, OpenCode, and availability via OpenRouter
• Free tier with 200 requests per month; paid plans start at $20/month with usage credits

Ideal Use Case

A team shipping a coding agent or AI IDE feature whose edit loop is slow and expensive because the main LLM rewrites whole files: route edits through Fast Apply so the frontier model only produces short edit snippets, and let Morph do the merge. The result is a visibly snappier product and a materially smaller inference bill.

How Morph differentiates

Morph is a Y Combinator company (S23) that focuses narrowly on inference optimized for code generation rather than general chat. Its published benchmarks — 10,500+ tokens per second apply speed at 98% accuracy, with a public coding-agent benchmark repo on GitHub — plus distribution through OpenRouter and community plugins for tools like Claude Code, OpenCode, Goose, and Crush have made "fast apply" a recognized pattern in coding-agent architecture.

FAQ

What is Morph? Morph is an API of specialized models for coding agents; its Fast Apply model merges AI-generated edits into files at 10,500+ tokens per second.

How does Fast Apply work? Your agent sends the original file, an abbreviated edit snippet with existing-code markers, and an instruction; Morph returns the complete merged file.

Is there a free tier? Yes — 200 requests per month free, with paid plans starting at $20/month including usage credits, plus per-token API pricing.

Can I use Morph with existing tools? Yes — via the API, OpenRouter, or plugins for Claude Code and OpenCode.

tl;dr

Morph (YC S23) sells fast, code-specialized inference — its Fast Apply model merges AI code edits at 10,500+ tokens per second, cutting agent latency and token costs versus full-file rewrites.

Related

Looking for more options? Browse the Developer Tools directory or read our best AI coding tools listicle. Morph is also tracked on Crunchbase.