Qwen Code or Claude Code: which terminal agent to pick
Two terminal agents with a similar feature set. We compare setup, models and API cost so the choice is not a blind one.
5 min · · сравнение · агенты
If you are choosing between Qwen Code and Claude Code, the question usually comes down to three things: what the agent itself can do, which models you can plug in, and what it costs when you pay per token. Below we compare Qwen Code vs Claude Code on exactly those points, with a calculation at Zukko prices as of October 18, 2026 and no claims about a "best" one.
What they have in common
Both tools start with a command in the project folder, read files, edit code, run commands and ask for confirmation on risky steps. The Qwen Code README contains a comparison table with Claude Code: according to it, both have subagents, hooks, MCP, a planning mode, a headless mode for scripts, an SDK and editor plugins. The table was written by the Qwen Code authors, so treat it as the project's own statement and verify the points you care about.
In daily work this means habits carry over: a project instructions file (CLAUDE.md for Claude Code, QWEN.md for Qwen Code, which also reads AGENTS.md), slash commands, MCP servers. If you have already written rules for one agent, the other takes little time.
Where they differ
| Item | Qwen Code | Claude Code |
|---|---|---|
| API protocols | OpenAI-compatible, Anthropic, Gemini, Qwen API, local servers | Anthropic format (/v1/messages) |
| Switching models | /model in a session, list defined in settings.json | ANTHROPIC_MODEL variable and model picker |
| Install | npm install -g @qwen-code/qwen-code@latest (Node.js 22+), install script and Homebrew available | see the documentation |
| Source code | open, per the project README | not marked as open in the same README table |
| Origin | a fork of Google Gemini CLI, now developed independently | an Anthropic product |
The main practical difference is the protocols. Qwen Code is built for many providers at once, so switching it to another model family is one line in a config. Claude Code speaks the Anthropic format by default. Through the Zukko gateway any model can be called in either format, so technically you can point Claude Code at a non-Claude model too. How well the agent works with it cannot be said in advance: test it on your own tasks.
Connecting both to one key
Create a fetch_... key in the Zukko dashboard. Claude Code uses the Anthropic address without /v1:
export ANTHROPIC_BASE_URL="https://api.zukko.pro"
export ANTHROPIC_AUTH_TOKEN="fetch_your_key"
export ANTHROPIC_MODEL="claude-sonnet-5-5"
claudeQwen Code in OpenAI mode takes the address with /v1:
export OPENAI_API_KEY="fetch_your_key"
export OPENAI_BASE_URL="https://api.zukko.pro/v1"
export OPENAI_MODEL="claude-sonnet-5-5"
qwenA detailed Qwen Code walkthrough with a persistent settings.json is in Qwen Code CLI: connecting. If you are unsure which address your client needs, see Anthropic or OpenAI format.
What one session costs
Take a hypothetical session: 1M tokens of fresh input and 100K tokens of output. Formula: input in millions × input price + output in millions × output price. Prices per 1M tokens as of October 18, 2026.
| Model | Zukko calculation | Total | For comparison: official prices |
|---|---|---|---|
claude-opus-5-5 | 1 × 0.8 + 0.1 × 4 | $1.20 | 1 × 4 + 0.1 × 20 = $6.00 |
claude-sonnet-5-5 | 1 × 0.4 + 0.1 × 2 | $0.60 | 1 × 2 + 0.1 × 10 = $3.00 |
gpt-6.1-sol | 1 × 0.27 + 0.1 × 1.32 | $0.402 | 1 × 2 + 0.1 × 10 = $3.00 |
qwen3.8-max | 1 × 0.088 + 0.1 × 0.238 | $0.1118 | 1 × 2 + 0.1 × 6 = $2.60 |
deepseek-v4.1-flash | 1 × 0.013 + 0.1 × 0.038 | $0.0168 | 1 × 0.15 + 0.1 × 0.6 = $0.21 |
A real agent session is rarely all fresh input: the agent resends the whole accumulated conversation at each step, and most of it comes from cache. For Claude the cache discount applies only in the Anthropic format, where the gateway sees cache reads and writes. Suppose that of 1M input, 100K is fresh, 850K is read from cache and 50K is written to cache, with 100K output. For claude-sonnet-5-5:
100K × 0.4 + 850K × 0.05 + 50K × 0.5 + 100K × 2 = 0.04 + 0.0425 + 0.025 + 0.2 = $0.3075.
The same volume in the OpenAI format (/v1/chat/completions) goes entirely at the input price for Claude, which is the $0.60 from the table. For GPT models cache reads are visible in both formats. So for Claude it is worth checking which format each agent really uses. Qwen Code also supports the Anthropic protocol, but check its documentation for which address it expects in that mode (with or without /v1): if requests go to /messages and you get a 404, add /v1.
More on agent session math: agent cost calculation with cache.
Access from Russia
Both agents call the API over HTTPS, and a key can come from any compatible provider. Access from Russia goes through the gateway: the dollar balance is topped up via SBP (the payment page adds 8.5% on top), MasterCard or cryptocurrency. You do not need a subscription to the official Claude or Qwen plans to work by key. Check the current subscription terms on the services' own sites.
How to choose
- You want one tool for several model families, including cheap ones: start with Qwen Code.
- All your work is built around Claude and you want the most native path for the Anthropic format: Claude Code.
- Not sure: install both, give them the same task and compare the result and the spend line in the "Usage" section of the dashboard.
Create a key in the dashboard, top up from $1, and the comparison takes half an hour.
FAQ
Qwen Code or OpenCode, which one?
That is another comparison of agents with many-provider support. OpenCode on Zukko needs the address with /v1, and Qwen Code in OpenAI mode does too. The difference is in the interface, feature set and feel, so give both the same task.
Claude Code or Codex?
In short: both connect to Zukko, differ in API format and habits, and the cost follows the table above.
Can I run Claude in Qwen Code?
Yes, via the OpenAI-compatible mode: put claude-sonnet-5-5 in OPENAI_MODEL. Remember the cache: for Claude the discount works in the Anthropic format.
Do I need a subscription?
No. API usage is paid per token from a single balance, and no Zukko subscription is required.
zukko
Pay 25×
less
for Claude Code & ChatGPT
One fetch_* key and base URL api.zukko.pro — for Cursor, Claude Code, Codex, and any SDK.
Related articles
- RAG or MCP: what is the difference and which to choose for giving a model dataRAG puts retrieved pieces into the prompt, MCP gives the model tools it calls itself. How they differ, what each costs and how to combine them.
- Claude Desktop or Claude Code: What Is the Difference and Which to PickThe desktop app and the console agent share one engine but sign in differently. We compare scenarios, model access and cost.
- opencode or Codex CLI: which terminal agent to chooseBoth agents are open source, both read AGENTS.md and both work with your own API. The differences are in providers, modes and cache billing. We compare and price a session.
- Function calling or MCP: what is the difference and what to chooseThese are not two rival methods but two levels of the same task. Who describes a tool, who runs it, and when you need MCP.