Unfortunately they jacked the price up to $50/month :( so this is old information already unfortunately!

I read that if you had bought it when it was at $3, it might still let you pay $10 or something, but this is unconfirmed.
I got rid of mine and am just using CliProxyAPIPlus to route Openrouter models to Claude Code.

But... I found this:

Rust redo of Claude Code, when the src code was leaked

Haven't tried it yet, and it might not have every detail yet from C.C. but it looks good, will try.

Will post new guide more like the older ones soon.

old info starts here

Alibaba Coding Plan: $3/Month for Claude Code

I had to post about this cheap Coding Plan I found — Alibaba Coding Plan, $3/month right now, and you get all these models plus some other ones:

Plan Details

Supported Models

Recommended:

• qwen3.5-plus (vision)
• kimi-k2.5 (vision)
• glm-5
• MiniMax-M2.5

More models:

• qwen3-max-2026-01-23
• qwen3-coder-next
• qwen3-coder-plus
• glm-4.7

You can use it directly in Claude Code, OpenClaw, and others.

I still use Claude Code and GitHub Copilot. I tried Goose... on Windows 11... errors and more errors! I don't have time for that so I got rid of it. I have noticed a lot of these things are created by Mac people, and Windows/Linux are 2nd class citizens — more of an afterthought.

Claude Code Setup for Alibaba / Dashscope

# --- ALIBABA / DASHSCOPE CLAUDE CODE CONFIGURATION ---

# PASTE YOUR API KEY HERE:
export ALIBABA_API_KEY="sk-sp-xxxxxxx"

# Internal helper function to launch Claude with Alibaba settings
_claude_ali_launch() {
    local model_name=$1
    shift
    (
        export ANTHROPIC_BASE_URL="https://coding-intl.dashscope.aliyuncs.com/apps/anthropic"
        export ANTHROPIC_AUTH_TOKEN="$ALIBABA_API_KEY"
        export ANTHROPIC_API_KEY=""  # Clear this to avoid conflicts
        export ANTHROPIC_MODEL="$model_name"
        export ANTHROPIC_DEFAULT_HAIKU_MODEL="$model_name"
        export ANTHROPIC_DEFAULT_OPUS_MODEL="$model_name"
        export ANTHROPIC_DEFAULT_SONNET_MODEL="$model_name"
        export ANTHROPIC_REASONING_MODEL="$model_name"

        # Performance/Token settings
        # export CLAUDE_CODE_MAX_OUTPUT_TOKENS=30000
        export API_TIMEOUT_MS=3000000
        export CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=1
        export CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1

        claude --dangerously-skip-permissions --model "$model_name" "$@"
    )
}

# Model Commands
alias claude-qwen='_claude_ali_launch qwen3.5-plus'
alias claude-kimi='_claude_ali_launch kimi-k2.5'
alias claude-glm5='_claude_ali_launch glm-5'
alias claude-mm='_claude_ali_launch MiniMax-M2.5'

# --- MENU DISPLAY ---
claude-menu() {
    echo -e "\n\033[1;36m====================================================\033[0m"
    echo -e "\033[1;32m  CLAUDE CODE - ALIBABA (DASHSCOPE) ENDPOINTS       \033[0m"
    echo -e "\033[1;36m====================================================\033[0m"
    echo -e "  \033[1;33mclaude-qwen\033[0m  - Use Qwen 3.5 Plus"
    echo -e "  \033[1;33mclaude-kimi\033[0m  - Use Kimi K2.5"
    echo -e "  \033[1;33mclaude-glm5\033[0m  - Use GLM-5"
    echo -e "  \033[1;33mclaude-mm\033[0m    - Use MiniMax M2.5"
    echo -e "\033[1;36m----------------------------------------------------\033[0m"
    echo -e "  Keys are pre-configured with skip-permissions."
    echo -e "\033[1;36m====================================================\033[0m\n"
}

# Run the menu automatically on startup
claude-menu

Which Model to Use?

I have found that running Claude Code set to kimi-k2.5 will create very elaborate, complex plans — but it is not as fast as the others. MiniMax M2.5 is quickest, which is often the best just for that reason.

Some tactics you can use:

• Run these different model commands, even at the same time, to all create plans simultaneously.
• Then have one of the smarter ones like Kimi go over all the plans and give opinions — explain why any one is better than the others.
• Generate a best-of-N plan. Use MiniMax to implement.

I haven't used the new Qwen 3.5 models enough yet to have a strong opinion, but they seem good. Same with GLM 5.

The Coding Plan's main page gives instructions for all of the main coding agentic tools, and they include OpenClaw. Save some money! Be careful with security if messing with OpenClaw. 🔐

Gemini 3.1 Pro came out too! It's very very good. The USA AI companies are still the absolute best ones. It still makes sense to use something like my tool, aicodeprep-gui, to plan things out on all the different web chat's (see the older guides from last year... July, Sept etc for how-to's) and just use that tool to go back and forth when bug fixing or in the planning stages of a project.

I use Gemini 3 Flash a lot for bug fixing, the speed is sometimes better than intelligence since a faster model can fail 3 times before fixing something, while the smarter model can figure it out on first pass or two, but it takes longer...