Mistral Large 4 entered public API preview on October 6; weights are planned for the end of October. The company describes native multimodal capability and specialist performance, with a one-million-token context window in its documentation. The launch discounts run for two weeks: $0.68 per million uncached input tokens and $2.09 per million output tokens. A hypothetical million-token input plus 10,000-token output would cost about $0.70 at those prices, before other operating costs. Release · Current model documentation.
Simon Willison tried his pelican-on-a-bicycle SVG prompt with reasoning set to none and high. He preferred the high version, which used fewer output tokens; the displayed drawing still put both pedals toward the front of the frame. One drawing says little about enterprise reliability, but it is an inspectable reminder that effort, token count and correctness need separate measurements. See the actual experiment.