Liquid AI's d1 goes live on OpenRouter at $0.04 per million input tokens
@OpenRouter announced on 1 October 2026 at 15:30 BRT that Liquid AI’s d1 is live as liquid/d1: “send your app’s state and yes/no, choice, or score questions, and get back typed answers with a probability for every option”. Terms: $0.04 / M input, $0 output, 65K context, and “Zero data retention” (OpenRouter’s wording; see the correction below). By the evening the post had ~160 likes and ~22k views.
That delivers the “available on OpenRouter soon” from d1’s launch on 29 September and gives the model its first price; Liquid’s own API still shows only the free d1:free tier. The price matches the $0.04 of Perplexity’s Decisions API launched the same afternoon, just under Jev’s $0.042.
How we list it. The d1 page gets the OpenRouter section and price. Maturity goes from 4.9 to 5.4 (integrations 3 → 5), adoption from 23 to 27, and capability stays at 6.5: there is still no latency, calibration figure or independent benchmark.
Vercel AI Gateway (9 October 2026)
Vercel added d1 to AI Gateway, with vision support, the Vercel AI SDK experimental_decide interface and OpenAI/TypeSafe-compatible routes. This is an integration update, not a new model or benchmark; catalog scores remain unchanged. Liquid announcement.
Correction and data-handling note (17 Oct 2026)
“Zero data retention” is OpenRouter’s wording in its 1 Oct post, not a statement from Liquid. On 17 Oct we re-read the sources: OpenRouter’s own liquid/d1 page makes no data-policy claim for d1. Its sidebar of other Liquid models does say that requests to LFM2.5-Embedding-350M “may be retained and used to train Liquid models”, and the same for LFM2.5-2.6B; those lines are about those models, not d1. Liquid’s Terms say it “may retain Your Content and Usage Data as reasonably necessary to provide, secure, and improve the Services” and that service-specific or third-party-platform terms apply where they exist; its privacy policy lists improving AI models with “inputs and outputs you share” among its purposes for personal data. Liquid’s pricing page says it never uses customer data to train models, in the enterprise section. Liquid’s d1 docs say nothing on retention. Net: the zero-retention claim for the OpenRouter route rests on one post; check before sending sensitive data.