Choosing and switching AI coding models

Darce works with 250+ coding models through OpenRouter: Claude, GPT, Gemini, Qwen, DeepSeek, Kimi, GLM, Grok and more. You don't need your own provider keys; requests go through your Darce plan.

The model picker

/model (or Ctrl+P) opens with your recent models, then OpenRouter's live most popular ranking, then everything else. Each row shows a friendly name, the vendor, a price level ($ to $$$$), the context size and what the model can do (vision, reasoning).

The picker lists models that can actually drive a coding agent: tool calling, a 64k+ context, current generation. Search matches every word you type and ranks results by popularity.

Switching models

/model sonnet-5.5        # switch by name
/model kimi              # search; switches if there's one match
darce --model openai/gpt-5.6-sol

Any model not shown in the picker still works by its full id.

Gear shift

Shift+↑ and Shift+↓ move to a smarter or cheaper model instantly, even in the middle of a task, with the price difference shown before the next step runs. The order comes from gears in ~/.darcerc:

{
  "gears": ["qwen/qwen3-coder-next", "qwen/qwen3-coder", "deepseek/deepseek-v4-pro", "anthropic/claude-sonnet-5.5"]
}

A sensible pattern: start cheap and fast for exploration, shift up for the hard part, shift back down for cleanup.

Which models each plan includes

Images and vision

• Paste a screenshot with Ctrl+V or drag image files into the terminal. If your current model can't see images, Darce hands that message to a fast vision model automatically (choose it with visionModel).
• Darce can generate images into your project (icons, illustrations, hero images, or edits of an existing image). The default is OpenAI's GPT Image 2.5, which can make transparent backgrounds; an image uses your allowance at its real cost, about a cent for a default one. Google's Nano Banana 2 and Pro (paid plans) and Seedream are available too.

Prompt caching

Each conversation stays on one model provider and marks its repeated prefix (instructions and tool list) for caching, so follow-up calls reuse it: cheaper and faster replies, especially with Claude, Gemini and Qwen.

Model directory

See the model directory for context windows, prices and the exact darce --model command for popular coding models. To compare models on your own task, use /derby.

Next: Commands · All docs · Model directory · Claude Code alternative