DeepSeek V3.2 as a coding agent in your terminal
DeepSeek V3.2 by DeepSeek, one of 250+ models you can use in Darce, the open source AI coding agent you can undo.
| Model | DeepSeek V3.2 |
|---|---|
| Vendor | DeepSeek |
| Model id | deepseek/deepseek-v3.2 |
| Context window | 164k tokens |
| Price per 1M tokens | $0.26 input · $0.80 output (OpenRouter) |
| Can do | tool calling · reasoning |
| Category | Open weights and budget |
Strengths
DeepSeek V3.2, a well-known, very inexpensive open model for everyday coding.
OpenRouter's description:
DeepSeek-V3.2 is a large language model designed to harmonize high computational efficiency with strong reasoning and agentic tool-use performance. It introduces DeepSeek Sparse Attention (DSA), a fine-grained sparse attention mechanism...
Use DeepSeek V3.2 as a coding agent in your terminal
Start Darce in your project with DeepSeek V3.2:
npm install -g darce-cli
darce --model deepseek/deepseek-v3.2Already in a session? Type /model deepseek/deepseek-v3.2, or open the picker with Ctrl+P. Darce then reads your code, edits files and runs commands with DeepSeek V3.2; risky commands ask first, and /undo reverses any change, including what shell commands did. No account yet? npx darce-cli lets you try it free. Which plan includes which models is on the pricing page.
Make it your default or a gear
In ~/.darcerc, set it as the starting model or add it to the gears you move between with Shift+↑ / Shift+↓ (see configuration):
{
"router": { "default": "deepseek/deepseek-v3.2" },
"gears": ["qwen/qwen3-coder", "deepseek/deepseek-v3.2"]
}Compare it on your own code
Benchmarks rarely match your repository. Race DeepSeek V3.2 against two other models on a real task and keep the best diff:
/derby --models deepseek/deepseek-v3.2,deepseek/deepseek-v4-pro,deepseek/deepseek-v4.1-flash fix the failing testRead more about /derby and /swarm.
Related models
- DeepSeek V4 Pro 0423 (DeepSeek): 1.05M tokens, $0.95 / $1.90 per 1M tokens
- DeepSeek V4.1 Flash (DeepSeek): 1.05M tokens, $0.30 / $1.20 per 1M tokens
- gpt-oss-120b (OpenAI): 131k tokens, $0.037 / $0.17 per 1M tokens
- Gemma 4 31B (Google): 262k tokens, $0.090 / $0.34 per 1M tokens
- Qwen3 Coder 30B A3B Instruct (Qwen (Alibaba)): 262k tokens, $0.070 / $0.28 per 1M tokens
- Kimi K2.6 (Moonshot AI): 262k tokens, $0.47 / $2.45 per 1M tokens
All models · Getting started · Claude Code alternative
Model data from OpenRouter, refreshed daily. Prices are OpenRouter's list rates and show relative cost; your Darce allowance is used in proportion to them.