On NVIDIA build.nvidia.com (NIM), DeepSeek-V4.1-Flash and GLM-5.3-Flash currently expose free OpenAI-compatible API endpoints. Set Base URL to https://integrate.api.nvidia.com/v1, add a Developer API Key, and wire common clients. Kimi K3.1 Pro is listed, but NVIDIA says API access is not available yet—use Playground only. Do not treat a working web chat as Agent-ready. Trial terms, credit-card rules, quotas, and rate limits follow the live pages and NVIDIA API Trial Terms.
| Model | ID | Free API? | What you can do now |
|---|---|---|---|
| DeepSeek-V4.1-Flash | deepseek-ai/deepseek-v4.1-flash | Yes | Base URL + Key + ID in Cursor/Codex-style clients |
| GLM-5.3-Flash | z-ai/glm-5.3-flash | Yes | Same; skip deprecated non-Flash z-ai/glm-5.3 |
| Kimi K3.1 Pro | moonshotai/kimi-k3.1-pro | No (not available yet) | Playground only; Key will not authorize API calls |
Register → verify (China/+86 SMS is offered; success not guaranteed) → avatar → API Keys → Generate → copy once. Base URL must include /v1. Do not copy Claude Code’s root-only ANTHROPIC_BASE_URL rule here. Paste model IDs as separate strings—never glue them. Smoke-test with a one-word reply before long jobs. Free tier is for evaluation; card-binding and rate limits are third-party/clause-checked, not promised forever.
Sources: https://build.nvidia.com/deepseek-ai/deepseek-v4.1-flash · https://build.nvidia.com/z-ai/glm-5.3-flash · https://build.nvidia.com/moonshotai/kimi-k3_1-pro · related Base URL tip https://new.nat6.net/news/custom-e410786ad6764d07/ · verification date 2026-11-07 (Asia/Shanghai).