https://nivana.one/blog Notes on model routing, what AI answers cost and how AI models compare, from the team building Nivana. en https://nivana.one/blog/why-ai-forgets-long-chats https://nivana.one/blog/why-ai-forgets-long-chats Tue, 27 Oct 2026 00:00:00 GMT Every model has a context window, and even inside it, details in the middle of a long conversation get lost. Here is what the research says about long contexts, and how Nivana summarises old turns so a long chat keeps its names, numbers and decisions. https://nivana.one/blog/why-ai-models-hallucinate https://nivana.one/blog/why-ai-models-hallucinate Tue, 27 Oct 2026 00:00:00 GMT AI models state wrong facts with total confidence. Research from OpenAI argues that training and testing reward guessing over admitting uncertainty. Here is why hallucinations happen, how they are measured, and what reduces them in practice. https://nivana.one/blog/what-happens-to-your-ai-chats https://nivana.one/blog/what-happens-to-your-ai-chats Tue, 27 Oct 2026 00:00:00 GMT When you type into an AI assistant, your words go to a model provider and often stay on someone's servers. Here is what zero data retention means, what Nivana stores and encrypts, and how private chats, the anonymiser and model controls work. https://nivana.one/blog/what-is-an-ai-model-router https://nivana.one/blog/what-is-an-ai-model-router Sat, 24 Oct 2026 00:00:00 GMT No single AI model is best at everything. A model router reads each request and sends it to the model most likely to answer it well, at the lowest cost that still gets the quality. Here is how routing works and what the research says it saves. https://nivana.one/blog/what-an-ai-answer-really-costs https://nivana.one/blog/what-an-ai-answer-really-costs Sat, 24 Oct 2026 00:00:00 GMT AI models are priced per million tokens, and the gap between the cheapest and the most expensive frontier model is now about seventy-fold. A plain guide to tokens, model prices in October 2026, and what that means for a chat, an image or a video. https://nivana.one/blog/how-hard-should-an-ai-think https://nivana.one/blog/how-hard-should-an-ai-think Sat, 24 Oct 2026 00:00:00 GMT Modern models can spend extra tokens thinking before they answer. That helps on maths and multi-step problems and is wasted on a greeting. What reasoning effort is, what it costs, and how Nivana decides it per message. https://nivana.one/blog/why-ai-answers-should-cite-sources https://nivana.one/blog/why-ai-answers-should-cite-sources Sat, 24 Oct 2026 00:00:00 GMT Models are trained on a snapshot and confidently fill gaps. Retrieval and citations reduce that, but recent studies show they do not remove it. What grounding is, what the evidence says, and how Nivana decides when to search. https://nivana.one/blog/best-ai-model-for-coding-writing-math https://nivana.one/blog/best-ai-model-for-coding-writing-math Sat, 24 Oct 2026 00:00:00 GMT Leaderboards rank models on an overall score, but the best model for your code is not the best for your essay. How public benchmarks differ, why one ranking misleads, and how Nivana measures quality per task to route by it. https://nivana.one/blog/ai-image-and-video-generation-models-compared https://nivana.one/blog/ai-image-and-video-generation-models-compared Sat, 24 Oct 2026 00:00:00 GMT Image and video models differ in what they can do, how long a clip can be, whether they make sound, and what a second costs. What the leading models offer in October 2026, with prices, and how Nivana picks one for each request.