Cursor + GLM-5.3 / Kimi K2.7 / DeepSeek: Custom Model Setup (2026)

Published 2026-09-13 · NovAI Team

Cursor is the most widely used AI code editor, and its paid plans meter fast requests against Anthropic/OpenAI. What fewer people know: Cursor lets you override the OpenAI API base URL — which means you can route its custom-model requests to NovAI and run Chinese frontier models on your own USD balance.

What works and what doesn't

• Works: any feature that uses custom OpenAI-compatible models — chat, Composer with custom models, and @-symbol context on your own key.
• Doesn't work: Cursor's built-in hosted models (their managed Claude/GPT endpoints) — those stay on Cursor's billing. This guide covers the custom-model path.

Setup

1. Open Settings → Models in Cursor.
2. Enable the OpenAI API Key section and paste your NovAI key (nvai-...).
3. Set Override OpenAI Base URL to https://aiapi-pro.com/v1.
4. Under Model Names, click + Add model and add the ids you want: glm-5.3, kimi-k2.7-code, deepseek-v4-pro, qwen3.8-max.
5. Disable Cursor's hosted models if you want every request on your own balance.

Practical tips

• Use glm-5.3 when a conversation needs the whole repo — its 1M context absorbs large codebases that would be truncated elsewhere.
• For Tab-completion-style latency sensitivity, prefer kimi-k2.7-code-highspeed ($2.073/$8.61) — the low-latency variant of Kimi's agent model.
• Token spend lands on your NovAI balance in real time; check usage in the dashboard.

Best NovAI models for coding agents

For long agent loops, kimi-k2.7-code and deepseek-v4-pro give the best cost-per-task; for repo-wide refactors that need the full codebase in context, glm-5.3's 1M window is unmatched. Use free glm-4.7-flash for planning and lightweight edits.

Frequently Asked Questions

Can Cursor use Chinese models like GLM or Kimi?

Yes, via the custom-model path: in Settings > Models, paste your NovAI API key, override the OpenAI Base URL to https://aiapi-pro.com/v1, and add model ids such as glm-5.3 or kimi-k2.7-code.

Does this replace my Cursor subscription?

Partially. Cursor's built-in hosted models stay on Cursor billing; requests to your custom OpenAI-compatible models are billed by NovAI at provider-list prices with zero platform fee.

Which NovAI model is best in Cursor?

glm-5.3 for repo-wide context (1M tokens), kimi-k2.7-code for edits and agent work, and kimi-k2.7-code-highspeed when you need the lowest latency.