Novita AI Llama 3.3 70B Instruct model page (serverless price and context)

Fetched 2026-10-17. Serverless card: Input $0.135 / M tokens, Output $0.4 / M tokens, model id meta-llama/llama-3.3-70b-instruct, quantization bf16, Context Length 12K, Max Output 12K. The sample request sets max_tokens=12288, which matches a 12K cap rather than the model's native 128K. Same prices as the 2026-10-14 OpenRouter Novita row and the 2026-10-17 pricing-page table. This is the primary source for the context correction: Novita exposes 12K on this serverless endpoint, not 131072.