How do I use Qwen 3.8 in Claude Code?
Set ANTHROPIC_BASE_URL to https://api.stdcmpt.com, a Standard Compute key in ANTHROPIC_AUTH_TOKEN and ANTHROPIC_MODEL to qwen/qwen3.8-max-0902: Claude Code then runs on Qwen 3.8 Max, with qwen/qwen3.8-flash for background calls, on a plan from $19 a month that stops new requests with HTTP 402 when its compute budget is spent. No Alibaba Cloud account is needed.
# Qwen 3.8 Max for the session, Qwen 3.8 Flash for subagents and background calls
# No /v1 here: Claude Code adds /v1/messages itself.
export ANTHROPIC_BASE_URL=https://api.stdcmpt.com
export ANTHROPIC_AUTH_TOKEN=YOUR_API_KEY
export ANTHROPIC_MODEL=qwen/qwen3.8-max-0902
export ANTHROPIC_DEFAULT_OPUS_MODEL=qwen/qwen3.8-max-0902
export ANTHROPIC_DEFAULT_SONNET_MODEL=qwen/qwen3.8-max-0902
export ANTHROPIC_DEFAULT_HAIKU_MODEL=qwen/qwen3.8-flash
export CLAUDE_CODE_SUBAGENT_MODEL=qwen/qwen3.8-flash
claude
Use the dated ID qwen/qwen3.8-max-0902; the undated name is a temporary alias. The Model Studio tab follows Alibaba's Claude Code guide for Token Plan subscribers.
Which Qwen ID goes where
| Model ID | Use it for | Reads images |
|---|---|---|
| qwen/qwen3.8-max-0902 | The main session and the opus slot. Reasoning from minimal to xhigh | Yes |
| qwen/qwen3.8-flash | Background calls and subagents. Reasoning on or off | Yes |
| qwen/qwen3.8-27b | A middle option for reading-heavy work | Yes, and video |
Use the dated Max ID. The undated qwen/qwen3.8-max is a temporary alias kept for older setups.
What works with Qwen in Claude Code
- Tools, MCP servers and pasted screenshots work on all three IDs.
- The thinking toggle does nothing on this endpoint; set Max's reasoning level on the Models page.
- The WebSearch tool is not available, because the endpoint drops Anthropic's server tools.
/modeldoes not list Qwen IDs; type the ID, as in/model qwen/qwen3.8-flash.
What a day of Qwen 3.8 Max costs
| Per day | Alibaba list price | Starter ($19 a month, $20 of compute) budget |
|---|---|---|
| Typical day | $4.30 | About 4 such days |
| Busy day | $44 | About half a Standard budget |
Model Studio, Singapore region, per million tokens: Qwen 3.8 Max $2 input, $0.25 implicit cache read, $6 output; Qwen 3.8 Flash $0.15 input, $0.47 output. Source. A typical day is 8.5M tokens and a busy one 86.7M, 87.2% of input as cache reads: the median and 90th-percentile coding-agent day in our Agent Burn Index, measured on OpenCode, Cline, Kilo and similar agents, September 13 to October 12, 2026. Priced at the list rates above; your own days will differ.
The catch
Alibaba runs its own Claude Code endpoint for Token Plan subscribers (https://token-plan.ap-southeast-1.maas.aliyuncs.com/apps/anthropic), mapping opus to qwen3.8-max, sonnet to qwen3.8-flash and haiku to qwen3.6-flash. If Qwen is the only family you use and you are happy with an Alibaba Cloud account, compare that plan first.
Where Standard Compute fits
No Alibaba Cloud account or region choice: Qwen 3.8 Max for the session, a Claude or GPT model one /model away, billed in US dollars.
FAQ
Can I use Qwen3-Coder in Claude Code here?
Not on Standard Compute: the catalog serves the Qwen 3.7 and 3.8 generations (Max, Plus, Flash, 27B), not the Coder line. Qwen 3.8 Max is the strongest Qwen model here.
Can I run Qwen locally with Claude Code instead?
Yes, through a local proxy such as LiteLLM in front of a local server, if your hardware can hold the model. This page covers the hosted route.