GreenPT / greenpt/pixtral-12b-2409

Pixtral 12B - access through LLMTR

Pixtral 12B is a small vision-language model with a 128,000-token context window. Input and output tokens bill at the same rate — one of two rows on this provider where that holds, the other being Llama 3.3 70B Instruct, which costs considerably more. Most rows charge two to six times more for output, so on image work that produces long descriptions this makes the total predictable. It genuinely processes image input: in measurement it described a product photo correctly and spent over 2,800 input tokens on the image — noticeably more than the other rows here spent on the same image. That measured difference is a cost difference; no image-quality comparison was run between the rows. It does not reason by default. Tool/function calling was verified over a full two-turn loop. Repeated prefixes are not served from a cache.