Claude Code too expensive? Keep the tool. Swap the engine.
Claude Code is free — what costs money is the model behind it. If Opus 4.8 at $25/1M output tokens is stretching your budget, you don't need a new tool: Claude Code can drive any Anthropic-protocol endpoint. Point it at a different engine and your workflow doesn't change.
What the same work costs
| Model | Output $/1M | vs Opus 4.8 |
|---|---|---|
| Claude Opus 4.8 | $25.00 | — |
| Claude Sonnet 5 | $10.00 (intro, rises Aug 31) | 2.5× cheaper |
| qwen3.8-max | $3.11 | 8× cheaper |
| glm-5.2 | $2.57 | 9× cheaper |
| qwen3.7-plus | $0.74 | 33× cheaper |
| deepseek-v4-flash | $0.18 | 138× cheaper |
These are China's frontier models — the official full-quality builds, which sit alongside US flagships on SWE-bench-class coding evals. Don't trust benchmarks: try them free on your own code and judge.
The two-line setup
export ANTHROPIC_BASE_URL="https://api.gloritoken.ai" export ANTHROPIC_AUTH_TOKEN="sk-your-key"
That's the whole migration. Run claude and you're coding on qwen3.8-max; switch models anytime with /model (kimi-k3, deepseek-v4-pro, glm-5.2…). Cline, Cursor and Roo Code work the same way via the OpenAI-compatible endpoint — full guides.
Prices August 2026 from public pricing pages; Alibaba-served models are 12–48% below the vendors' international rates (Qwen 3.8 Max: 48% below; Kimi K3: 42% below), cached input discounted at the same ratio. gloritoken is an independent platform, not affiliated with Anthropic; we do not relay Claude, GPT or Gemini — we serve China's frontier models first-hand.