NewHow we catch a seller listing the wrong model, and refund you if one does.
deepseek logo

How do I use DeepSeek V3.2 in OpenCode?

deepseek/deepseek-v3.2

OpenCode reaches DeepSeek V3.2 through the OpenAI-compatible chat address of MartLLM. In ~/.config/opencode/opencode.json, add a provider named uberllm that uses the @ai-sdk/openai-compatible package with baseURL set to https://api.martllm.com/v1, list deepseek/deepseek-v3.2 under its models, put your MartLLM key in the UBERLLM_API_KEY environment variable, then pick uberllm/deepseek/deepseek-v3.2 with /models. The limit numbers are the context size and the largest reply of DeepSeek V3.2; OpenCode uses them to know when to compact a long session.

json
{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "uberllm": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "MartLLM",
      "options": {
        "baseURL": "https://api.martllm.com/v1",
        "apiKey": "{env:UBERLLM_API_KEY}"
      },
      "models": {
        "deepseek/deepseek-v3.2": {
          "name": "DeepSeek V3.2",
          "limit": {
            "context": 160000,
            "output": 32768
          }
        }
      }
    }
  },
  "model": "uberllm/deepseek/deepseek-v3.2"
}

OpenCode calls: OpenAI chat completions https://api.martllm.com/v1/chat/completions · Full OpenCode guide

Replace ull_your_key with your MartLLM API key. One key works in every tool and for every model.

By API, DeepSeek V3.2 answers on three addresses with the same MartLLM key and the model id deepseek/deepseek-v3.2: OpenAI chat completions at https://api.martllm.com/v1/chat/completions, the OpenAI Responses API at https://api.martllm.com/v1/responses and the Anthropic Messages API at https://api.martllm.com/v1/messages.

How much does the DeepSeek V3.2 API cost?

On November 20, 2026, DeepSeek V3.2 costs $0.0026 per million input tokens and $0.0038 per million output tokens through MartLLM, 99% below its list price of $0.26 and $0.40. 60 sellers on the market can serve it. In the 7 days before, half of its 3,363 streamed requests got their first token within 5.4 seconds, with a typical prompt of 47,976 tokens. It reads up to 160,000 tokens of context. MartLLM adds no fee to these prices, and a request that fails is not billed.

Prices as of 11:00 UTC.

DeepSeek V3.2 in other tools

Other models in OpenCode