Amazon: Nova Premier 1.0

Amazon Nova Premier 1.0 is a text generation model from Amazon, listed on api14 under the model id amazon/nova-premier-v1. It sits in the text category and is described by Amazon as the most capable of its multimodal models for complex reasoning tasks, and as a strong teacher for distilling custom models. If you are already working within the Amazon ecosystem or want a reasoning focused model that can also serve as a distillation source for smaller custom models, this is the variant to evaluate first.

The model accepts text and image input and returns text output, so you can mix prompts that contain written instructions with images in the same request. It has a context length of 1,000,000 tokens and a maximum output of 32,000 tokens. That combination is useful when you need to feed in very long documents, large code files or long conversation histories while still leaving room for a substantial generated response. The context window and output ceiling are the two numbers that most directly shape how you structure prompts, since both are generous relative to typical text models.

Because it handles images alongside text and is aimed at complex reasoning, natural fits include document understanding with embedded figures, chart or diagram interpretation where the answer requires reasoning over what the image shows, and multi step analysis over long inputs. The large context window also suits whole repository or whole report summarisation, retrieval augmented generation where many chunks are supplied at once, and agent style workflows that accumulate a long history before producing a final answer. The distillation angle matters too: if you plan to train a smaller custom model, Nova Premier can act as the teacher generating the training signal.

You call it through api14's OpenAI-compatible API using the model id amazon/nova-premier-v1. Supported request parameters are max_tokens, stop, temperature, tools, top_k and top_p. The presence of tools means you can wire it into function or tool calling flows, and stop sequences let you control where generation halts. Sampling behaviour is tunable through temperature, top_p and top_k together. Set max_tokens with the 32,000 output ceiling in mind, and remember that input can include images as well as text.

There is no listed limitation beyond the output cap and the fact that output is text only, so if you need generated images or audio this model will not provide them. Since the provider describes it as the most capable in the family, it is the natural pick when reasoning quality matters more than latency or throughput, and a smaller sibling may fit better for high volume, simpler text tasks. Release date listed by the provider is 2025-10-31.