Claude Code setup

Install the Agent using its official guide below. Terminal examples use macOS / Linux or Windows WSL. If already installed, check your version and continue.

Use Claude Code 2.1.242 or later for the eight-model picker in this guide. Run claude --version and update through the official installation guide if needed. Older versions can still select each model with the exact /model commands below.

Have your delivered Multi-Model Gateway key ready. You will insert it in the local settings file in the next step; there is no separate shell export.

This guide requires activated access and the confirmed gateway URL supplied with your key.

Create ~/.config/multi-model-gateway/claude and save settings.json there. On Windows, follow these commands inside WSL. Replace the key placeholder locally and keep the file private.

This is a separate profile for the gateway. Keep your normal Claude settings as they are. If this gateway file already exists, merge the fields into it.

Download settings.json below or copy the identical JSON. It includes the gateway connection and all eight modelPicker entries, so /model lists the actual model names. Replace the key placeholder before starting Claude Code.

Read the three parts separately: model selects the startup model; modelPicker.options lists all eight selectable models; env supplies the connection and alias mappings. The three ANTHROPIC_DEFAULT_*_MODEL values are fixed shortcut targets, not the model list. This example points them all to DeepSeek Flash so calls using Claude aliases reach an available gateway model. You may assign each a different gateway ID. These mappings do not follow later /model changes.

Save this as the user settings file in the dedicated directory shown here, not as a project's .claude/settings.json: Claude ignores modelPicker in project settings. If the menu is unchanged, check your version and /status for managed settings that override the lineup.

No model renaming is needed in this CLI setup. Open /model to see all eight models by their real names, then select the one you want. Each modelPicker entry pairs a readable label with the exact gateway model ID.

The settings in step 3 replace the built-in picker lineup with the eight gateway models. Claude also keeps a Default row, which can add a context suffix such as [1m]; select a named gateway row to use its exact ID.

The opus, sonnet and haiku aliases have separate, fixed mappings. In this example, /model opus selects DeepSeek Flash, while /model kimi-k3 selects Kimi K3. Switching the main model does not update those mappings. The aliases also apply to opusplan, agents that explicitly request them, and Haiku background tasks. You can change each ANTHROPIC_DEFAULT_*_MODEL value to any gateway ID listed above, then restart.

You can also type an exact command below, such as /model minimax-m3. This works as a fallback on versions older than 2.1.242. Desktop uses a separate compatibility endpoint and Claude-style IDs; do not copy its IDs into this CLI profile.

In a trusted project folder, run the command below each time you want a gateway session. CLAUDE_CONFIG_DIR separates its user settings, sessions and plugins from your normal Claude profile.

Run /status to confirm the endpoint and setting sources. Project and managed settings can still apply; inspect them if the provider differs. Use the exact /model commands from step 4 to switch. This launch command starts a fresh session on DeepSeek Flash each time.

Start on DeepSeek Flash. Choose Low, send a short file-reading task, then choose High and repeat. Check the current setting after each change. A longer answer by itself does not prove the effort changed.

Effort means how much reasoning to request. Kimi K3, GLM 5.3 and GLM 5.3 Flash offer Low / High / Max with thinking always enabled. Choose Low for shorter reasoning; there is no Off option. Their model default is Max, but a client setting can override it. DeepSeek and Qwen have their own intensity levels listed below.

MiMo has an On / Off switch; High is the client's label for On. MiniMax M3 supports Off / Auto / On through thinking.type (disabled / adaptive / enabled). Its effort-only menus expose Off and High (On); Auto needs a native thinking control. These modes are not separate Low / High / Max intensities. Follow this Agent's default and switching instructions below; a supported control does not set a session's starting value.

Run /effort status to inspect the effective setting. /effort opens the selector; /effort low and /effort high set a level. In current versions these choices can save a per-model default in modelSettings. Use the selector's session-only action or a launch flag such as --effort high for a temporary choice; Max is session-only. Use /effort auto to clear your manual selection; Claude can still choose and send its own effort.

For MiniMax M3, start a dedicated CLI session with the native mode below. Replace adaptive with enabled for On. For Off, use the second command, which also clears the conflicting effort. This launch setting applies to every request in that session; exit and relaunch normally before switching to another model.

Open /model and select DeepSeek Flash, then MiniMax M3. The downloaded settings list all eight models. You can also enter the exact /model commands below. No key or URL changes are needed.

Open a project with a README.md and send the file-reading prompt below. Approve access to the file if prompted. You can switch models later without changing the key or gateway address.

You can also ask a question or search for current documentation. Send each example as a separate message.