Decisions at the
speed of light.

System 1 thinking is fast and intuitive; System 2 is slow and deliberate. A System 1.5 model moves between the two: it answers most questions in a single fast pass and thinks only when a question needs it, and either way it returns a typed decision. Wity is the first System 1.5 model, built for decisions inside software.

Five ways. It's multimodal, so it reads images, screenshots and scans as well as text. It reads long inputs, with a 128K-token context window, so a whole contract or thread fits in one request. It adds a generate primitive for the short text an action needs. It thinks automatically when an answer is uncertain. And it can run on your own premises.

Text, images, screenshots and scanned pages, mixed however you like in one request, within a 128K-token context window. Every question is answered against all of it.

Up to 128K tokens: a full contract, a long email thread or a stack of scanned pages. Many decision models take only a short passage, which means chunking the input and hoping the part that matters survives. With Wity you send the whole thing, and every question is answered against all of it.

generate writes short text for an action: a value to fill in, a city to type, a one-line reply. It runs on the same state right after a decision, and its output follows a shape you define, so you can validate it like any other input. Decisions stay typed; generate covers the few places where software genuinely needs words.

With reasoning set to "auto", Wity answers directly when it is clear and thinks first when an answer is uncertain or unstable. You can also turn thinking off for the lowest latency, or always on. The output type is the same in every mode.

Yes. We deploy Wity on your premises, inside your network, with the same API and SDKs as Wity Cloud. Your data never leaves your environment, and we handle installation, tuning and updates with your team.

Structured outputs constrain the format of an LLM's text, but the model is still writing prose underneath with no sense of how sure it is. Wity's answer is a typed decision by construction, with a confidence attached.

We ran Wity and eight fast, low-cost models, among them Gemini 3.5 Flash-Lite, Claude Haiku 4.5, GPT-6 Luna at low reasoning and DeepSeek V4.1 Flash, through the same 200 real-world cases. Wity's typical call took 109 ms, against 943 ms for the next fastest, and 1,000 runs cost $3.29, against $17.95 for the next cheapest. Averaged over the eight, Wity was 22.8× faster in model time and 24.8× cheaper.

Image JevBench, run independently by Benchmark Heaven, put 50 systems through 684 image decisions, 456 of them sealed. Wity-1 placed first on the official composite at 80.20, ahead of Imajev-4B at 76.39, with 89.9% accuracy on sealed items and 87.7% on public ones. It also leads the 38 Jev-class systems on Capability Score, at 85.7.

Yes. No model is perfect. Use the confidence on each decision to decide when to act and when to escalate, and validate generated text before it touches anything important, exactly as you would with any other input.

Create an account to get an API key, read the docs, or join our Discord. For on-prem deployments and enterprise questions, email wity@alphanimble.com.