GroqCloud vs Nemotron 3.5 Super VL

GroqCloud

Free plan with no card, at 30 requests a minute and 1,000 a day on gpt-oss. Four model shutdown dates between 2026-07-17 and 2026-09-21, with no stated minimum notice.

Nemotron 3.5 Super VL

NVIDIA's free endpoint takes OpenAI-format requests with typed tool calls, JSON Schema output and an OpenAPI 3.1 file, and the weights run anywhere under a permissive licence. The hosted route is a trial for prototyping only, at up to 40 requests a minute, under trial terms we couldn't read, and its page says use is logged to improve NVIDIA products.

GroqCloud

1. Call /models at start-up. Four model ids stopped working between July and September
2. Read retry-after on a 429 and the x-ratelimit-remaining-tokens header before the next call
3. Free plan allows 8,000 tokens a minute on gpt-oss, so keep prompts small or batch them
4. Don't build on Qwen 3.8 27B. It's a preview and previews can go at short notice
5. Treat a 498 as Flex capacity and retry later; server errors aren't billed

Nemotron 3.5 Super VL

1. Set chat_template_kwargs to {"enable_thinking": false} for short tasks. Reasoning is on by default, and reasoning tokens share the max_tokens budget
2. Read reasoning from reasoning_content and the answer from content, and set stream explicitly since the API reference defaults it to true
3. Don't send confidential or personal data to the hosted trial. The model page asks callers not to and says use is logged
4. Back off and retry on throttling. Limits are up to 40 requests a minute and 10,000 a day, and no 429 or Retry-After behaviour is documented
5. For production, use a partner endpoint or serve the weights. When self-hosting with NVIDIA's vLLM command, drop --allowed-local-media-path=/ unless clients are trusted