Current as of April 2026. Qwen has evolved into a top-tier family for developers, specifically with their Coder variants that rival proprietary models. In OpenCode, these models excel due to strong tool-calling capabilities and high context limits, making them ideal for multi-file refactoring and CJK-heavy documentation.

The quick answer

ModelInput / OutputContextBest For
Qwen3 235B A22B$0.07 / $0.10262KThe Budget Reasoning Engine
Qwen3.5-Flash$0.07 / $0.261MThe Massive Context Bargain
Qwen3.5-35B-A3B$0.16 / $1.30262KEfficiency for Simple Scripts
Qwen3.5-27B$0.20 / $1.56262KDense Stability for Tool Calls
Qwen3 Coder$0.22 / $1.00262KThe Long-Form Generator
Qwen3.5-122B-A10B$0.26 / $2.08262KBalanced High-End Logic
Qwen3.5 397B A17B$0.39 / $2.34262KThe Multimodal Generalist
Qwen3 Coder Plus$0.65 / $3.251MThe Long-Form Generator

Start with Qwen3 Coder Plus unless you have a specific reason to pick another. It is the only model in the family that combines a 1M context window with specialized coding weights. While it costs $0.65/M input, its ability to reason across an entire repository without losing the thread justifies the premium over the cheaper MoE variants.

Qwen3 235B A22B — The Budget Reasoning Engine

This massive MoE model is the cheapest way to get high-parameter reasoning at $0.07/M input. It is better for high-level architectural logic than the Flash model, though its 8K output limit makes it frustrating for generating long code files compared to the Coder series.

Qwen3.5-Flash — The Massive Context Bargain

At $0.07/M input and $0.26/M output, this is the go-to for RAG-heavy tasks. The 1M context window allows you to pipe massive amounts of documentation into OpenCode, though its tool-calling reliability is lower than the Coder Plus.

Qwen3.5-35B-A3B — Efficiency for Simple Scripts

This model is nearly identical to the 27B dense version in performance but uses an MoE architecture for faster throughput. Prefer this for quick unit test generation where you want lower latency and slightly lower input costs ($0.16/M) than the dense 27B.

Qwen3.5-27B — Dense Stability for Tool Calls

Unlike its MoE siblings, this dense model provides more consistent formatting for function calls. If OpenCode is struggling to parse JSON arguments from the 35B-A3B, use this model to stabilize your automated workflows.

Qwen3 Coder — The Long-Form Generator

This model features a massive 262K output cap, the highest in the family. It is the best choice for generating entire boilerplate projects or massive refactors in a single pass, and it is significantly cheaper than the Plus version at $0.22/M input.

Qwen3.5-122B-A10B — Balanced High-End Logic

This MoE hits the sweet spot for complex debugging. It offers significantly better reasoning than the 35B model for a modest price increase ($0.26/M input), making it a solid middle ground between budget models and the 397B flagship.

Qwen3.5 397B A17B — The Multimodal Generalist

Choose this if your coding task requires vision support, such as converting UI screenshots to code. It is the smartest general-purpose model in the family, though for pure text-based coding, the Coder Plus remains more precise.

Qwen3 Coder Plus — The Long-Form Generator

This model features a massive 262K output cap, the highest in the family. It is the best choice for generating entire boilerplate projects or massive refactors in a single pass, and it is significantly cheaper than the Plus version at $0.22/M input.

Setup in OpenCode

To use Qwen in OpenCode, add your provider (e.g., OpenRouter or Alibaba) to ~/.config/opencode/opencode.jsonc. Ensure the ‘provider’ field uses @ai-sdk/openai-compatible and set the ‘apiKey’ in ~/.local/share/opencode/auth.json. You must specify the full model string like ‘qwen/qwen3-coder-plus’ to route correctly.

Running through haimaker.ai

All Qwen models are also available through haimaker.ai. Wire haimaker as a single OpenAI-compatible provider and you get Qwen alongside every other frontier model:

{
  "provider": {
    "haimaker": {
      "npm": "@ai-sdk/openai-compatible",
      "options": {
        "baseURL": "https://api.haimaker.ai/v1"
      }
    }
  }
}

Direct provider setup

OpenCode ships with a built-in preset for Qwen (Alibaba). You do not need to configure a custom provider — just drop your API key into ~/.local/share/opencode/auth.json:

{
  "qwen": {
    "type": "api",
    "key": "your-qwen-api-key"
  }
}

Restart OpenCode and Qwen (Alibaba) models appear under /models. For providers not in the built-in directory (or to hit them through a gateway like haimaker), see the custom provider guide.

Bottom line

For serious development, use Qwen3 Coder Plus for its 1M context and coding specialization. If you are on a budget or just need to search through docs, Qwen3.5-Flash is the most cost-effective way to handle large codebases.

USE QWEN IN OPENCODE WITH HAIMAKER

See our OpenCode custom provider guide. See our Haimaker + OpenCode setup.