Multi-Model Gateway · Preview

One key. Your choice of coding models.

Start with DeepSeek, then switch to GLM, MiMo, Kimi, Qwen or MiniMax inside your Agent. Compare model descriptions and points rates below, then configure your tools once.

Model guide

Compare models and points rates

The plan price stays the same when you switch models. The points used per token change. Start with DeepSeek Flash at 1×, or compare other models using the same allowance.

Model IDs, points per token, points used by one million tokens and tokens available for one million points
Model / API ID	Points / token	1M tokens use	1M points provide · approx.
DeepSeek Flash`deepseek-flash` The default for getting started. One point per token, with the same shared allowance as every other model.	1×	1M points	1,000,000 tokens
GLM-5.3-Flash`glm-5.3-flash` GLM's Flash option at the same 1× rate as DeepSeek. Compare both on your own coding tasks.	1×	1M points	1,000,000 tokens
MiMo V2.6 Flash`mimo-flash` The lowest points rate in this lineup. The same balance provides about 43% more tokens than a 1× model.	0.7×	0.7M points	1,428,571 tokens
MiMo V2.6 Pro`mimo-pro` MiMo's Pro option. Uses twice the points of a 1× model and shares the same key as MiMo Flash.	2×	2M points	500,000 tokens
GLM-5.3`glm-5.3` The higher-rate GLM option. Each token uses nine times the points of GLM Flash.	9×	9M points	111,111 tokens
Kimi-K3`kimi-k3` The Kimi option in the shared catalog. At 20×, check your remaining allowance before a long session.	20×	20M points	50,000 tokens
Qwen3.8 Max`qwen3.8-max` Qwen's Max option at 15 points per token. Switch to it within your Agent using the same gateway key.	15×	15M points	66,666 tokens
MiniMax M3`minimax-m3` MiniMax M3 uses 5 points per token. Switch to it inside your Agent with the same gateway key and shared balance.	5×	5M points	200,000 tokens

Points used = total input and output tokens × model rate. Tokens = points ÷ model rate, rounded down to whole tokens. M = 1,000,000.

How usage is counted

Points = (input + output tokens) × model rate. For example, 1M DeepSeek Flash tokens plus 1M MiMo Pro tokens use 3M points from each allowance window.

Cached input counts once within input tokens; reasoning counts once within output tokens. These are shared-points rates, not separate per-token dollar charges.

Configuration guides

Connect your coding tool

Choose your Agent for a step-by-step article with terminal commands, configuration examples and expected results. Configure the gateway once, then switch models inside the Agent. Screenshots accompany the relevant setup steps.

Claude Code

Anthropic Messages

CLI setup

Claude Desktop · Code tab

Codex

OpenAI Responses

CLI setup

Codex Desktop · macOS / Windows

DSH (DeepSeek Harness)

OpenAI Chat Completions

CLI setup

Harness Desktop · macOS / Windows

Pi Coding Agent

OpenAI Chat Completions

CLI setup

Desktop terminal workflow

OpenCode

OpenAI Chat Completions

CLI setup

OpenCode Desktop

Hermes Agent

OpenAI Chat Completions

CLI setup

Hermes Desktop

OpenClaw

OpenAI Chat Completions

CLI setup

Control UI in your browser

What you need to connect

1. Your Agent
Install your coding tool. Its guide above includes the official download and the right configuration path.
2. Your gateway key
After activation, use the Multi-Model key and confirmed URL delivered to you. Other Coding Plan keys have a different scope.
3. One configuration
Save the provider and model list once. Choose models in the Agent and verify a short file-reading task.

Get help with setup