Unlock Claude Code Tokens and Access GLM‑5.1 with iFlytek’s Ultra‑Low‑Cost Coding Plan

Pricing and tiers

Coding Plan is a monthly AI‑coding subscription that bundles domestic models (星火 X2, GLM, Kimi, MiniMax, DeepSeek, Qwen) and supports both OpenAI and Anthropic protocols. Billing is per request count, eliminating token‑based cost concerns.

Carefree tier – 3.9 CNY one‑time purchase, unlimited requests. Includes Qwen3.5‑35B‑A3B, DeepSeek‑V3.2 and GLM‑4.7‑Flash.

Professional tier – adds GLM‑5, MiniMax‑M2.5 and Kimi‑K2.5 for stronger reasoning and longer context handling.

Efficient tier – provides GLM‑5.1, the strongest domestic coding model; GLM‑5.1 is available only in this tier.

Integration guide

Obtain API key

After purchasing a Coding Plan on the iFlytek MaaS platform, copy the dedicated API key from the subscription page:

https://maas.xfyun.cn/modelSquare?ch=MaaS-xbkol-JavaGuide

Base URL configuration

Use the dedicated coding endpoints:

OpenAI protocol – https://maas-coding-api.cn-huabei-1.xf-yun.com/v1 Anthropic protocol – https://maas-coding-api.cn-huabei-1.xf-yun.com/anthropic Common mistake: using the generic maas-api address instead of the required maas-coding-api prefix.

Model name

All requests should specify the unified model identifier: astron-code-latest The underlying model (e.g., GLM‑5.1) can be switched via the “Configure Model” button on the subscription page; the change typically takes 1–3 minutes to become effective.

Claude Code integration

{
  "env": {
    "ANTHROPIC_AUTH_TOKEN": "Your Coding Plan API Key",
    "ANTHROPIC_BASE_URL": "https://maas-coding-api.cn-huabei-1.xf-yun.com/anthropic",
    "CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC": 1,
    "API_TIMEOUT_MS": 600000,
    "ANTHROPIC_MODEL": "astron-code-latest",
    "ANTHROPIC_SMALL_FAST_MODEL": "astron-code-latest"
  },
  "permissions": {"allow": [], "deny": []}
}

Optional ~/.claude.json can contain:

{"hasCompletedOnboarding": true}

OpenClaw integration

{
  "provider": {
    "AstronCodingPlan": {
      "name": "讯飞星辰 Coding Plan",
      "options": {
        "baseURL": "https://maas-coding-api.cn-huabei-1.xf-yun.com/v1",
        "apiKey": "Your Coding Plan API Key"
      },
      "models": {
        "astron-code-latest": {"name": "Astron Coding Plan"}
      }
    }
  }
}

Cursor integration

In Cursor settings, add a new model with:

Override OpenAI Base URL – https://maas-coding-api.cn-huabei-1.xf-yun.com/v1 OpenAI API Key – your Coding Plan API Key

Model –

astron-code-latest

Model tier selection

Carefree tier uses Qwen3.5‑35B‑A3B, DeepSeek‑V3.2 and GLM‑4.7‑Flash. It is suited for high‑frequency, low‑cost tasks such as error lookup, code explanation, small function generation, commit messages, README drafts, and markdown polishing.

Professional tier adds GLM‑5, MiniMax‑M2.5 and Kimi‑K2.5, offering stronger reasoning and better handling of longer contexts, making it appropriate for architecture discussions, multi‑step refactoring, and larger code‑base analysis.

Efficient tier provides GLM‑5.1, which the author rates as the top domestic coding model. This tier is required for users whose primary goal is to leverage GLM‑5.1.

Practical use cases

Automated draw.io diagram generation

A draw.io skill built from https://github.com/Agents365-ai/drawio-skill generates architecture or flow diagrams for JavaGuide chapters via the /drawio-chart command. The skill reads section titles and key paragraphs, then produces professional‑looking diagrams.

Chinese markdown normalization

The chinese-markdown-normalizer skill scans a repository and fixes title hierarchy, spacing, punctuation, list indentation, and code‑block style. In a test run, Cursor executed six parallel agents on 70 files, detecting and correcting numerous formatting issues quickly.

Summary

The Carefree tier’s 3.9 CNY entry price offers a virtually risk‑free way to try AI‑coding assistance. Higher tiers provide stronger models for more demanding tasks. Users should monitor daily token caps and QPS limits when scaling usage.