Introduction

Open-weight coding models on an OpenAI-compatible API

Morph serves open-weight models on inference kernels built for coding agents. Same weights, a different engine. How →

Run an open-weight model

Point any OpenAI SDK at https://api.morphllm.com/v1. Existing code only needs a new model string. The Anthropic SDK works too, at /v1/messages (see Endpoints).

TypeScript

import OpenAI from 'openai';

const openai = new OpenAI({
  apiKey: process.env.MORPH_API_KEY,
  baseURL: 'https://api.morphllm.com/v1',
});

const chat = await openai.chat.completions.create({
  model: 'morph-glm53-744b',
  messages: [{ role: 'user', content: 'Write a rate limiter in TypeScript.' }],
});

console.log(chat.choices[0].message.content);

Models

Model	Model ID	Context
Kimi K3 2.8T	morph-kimik3	1M
GLM-5.3 744B	morph-glm53-744b	1M
GLM-5.3-Flash multimodal	morph-glm53flash	1M
DeepSeek V4.1 Flash multimodal	morph-dsv41flash	1M
DeepSeek V4.1 Flash 1017 multimodal	morph-dsv41flash-1017	1M
DeepSeek V4 Flash 0731	morph-dsv4flash	1M

Every model supports tool calls, JSON mode and JSON schema, logprobs, and reasoning. Per-token rates are on the pricing page and live at /api/models/json. DeepSeek V4.1 Flash Abliterated is served behind approval.

Built for agent loops

Three things change when the caller is an agent rather than a chat window.

• Prefix caching is on for every model with no configuration. Set an x-session-id header and each turn lands on the worker already holding its prefix.
• Agent programs: tag turns with a run_id and the scheduler pins the run to its KV cache, resumes it first after a tool call, and admits the whole run under load.
• Standby: service_tier: "standby" runs on spare capacity at half price. Sheds a fast 429 when a region is busy. Built for crons, batch jobs, and evals.

Speed comes from a speculator trained on each model’s own coding output (3.07x vs. 1.93x for a generic draft) and FP4 kernels tuned per GPU. How we optimize for codegen →

Specialized models

Small models that take one job off the main agent. Same key, same base URL.

Model	What it does	Speed
Fast Apply	Merges a lazy edit snippet (// ... existing code ...) into the file	10,500 tok/s, 98% accuracy
WarpGrep	Searches a repo in its own context window, returns file/line spans	~3.8 steps, #1 SWE-Bench Pro
Compact	Drops irrelevant lines from chat history, survivors verbatim	33,000 tok/s, 50-70% smaller
Router	Returns which model tier a prompt needs	~180ms, $0.005/request

Fast Apply and WarpGrep also ship as an MCP server (edit_file, codebase_search) for Claude Code, Cursor, Codex, and VS Code:

Terminal

npx -y @morphllm/morph-setup --morph-api-key YOUR_API_KEY

Dedicated endpoints

Reserve capacity for one model on its own pool. Hourly plans, your own endpoint URL, same OpenAI client. Dedicated endpoints →

Next steps

Open Source Models

Full model guide: Anthropic SDK, tool calls, structured outputs, approval-gated models

Prefix Caching

Session keys, cache-read pricing, hit-rate tips

Standby Requests

Half price for latency-tolerant work

Claude Code

Run Claude Code on Morph models

Fast Apply Quickstart

Prompt templates, tool definitions, MCP

API Playground

Try every model with no code

Enterprise

Dedicated instances, self-hosted deployments, zero data retention. 99.9% uptime SLA, SOC2, SSO.

Talk to Sales

Custom deployments and volume pricing