Morph

Morph is an inference provider built for coding agents. It offers an OpenAI compatible API that serves open weight models tuned for agent workloads, along with a set of specialized models for the steps an agent repeats. WarpGrep handles code search, Fast Apply merges code edits at a stated rate of more than 10,000 tokens per second, and Compact handles context compaction. Reflex is a fast model used as a judge on agent conversation turns, intended to catch failures that traces miss. It is aimed at teams building coding agents and developer tools that need low latency on edit application and search. The vendor says its hosted models run on kernels tuned for agents and that the lineup includes models from families such as GLM, Kimi, Qwen, MiniMax and DeepSeek, with context windows up to 1M tokens on some. It frames the choice as fast inference where a user is waiting and the lowest cost per token for background work. Access is by API key, through an MCP install, or through OpenRouter, and documentation and a playground are provided. Billing is per token. A blog post on the site lists $0.12 per million input tokens, $0.008 per million cached tokens and $0.80 per million output tokens for one recently announced model, and other models and specialized tools carry their own rates. The vendor also offers a call booking route for larger deployments.