AI that shows its work.

A family of AI models for math, science and code. It reasons step by step, checks itself, and tells you how it got there.

Request access Explore the models

The figure is a real neural network, learning in your browser. Click it to give it a point it gets wrong.

Every answer comes with its working.

Rosenblatt sets out each step and ends with a check, so you can follow the argument and catch a mistake, not just copy a result.

What do the first n odd numbers add up to?

Rosenblatt

Try small cases: 1 = 1, 1 + 3 = 4, 1 + 3 + 5 = 9. They look like squares, so guess n².
Prove it by induction. For n = 1 the sum is 1 = 1².
Suppose the first n odd numbers sum to n². The next one is 2n + 1.
Then the first n + 1 sum to n² + 2n + 1 = (n + 1)², so the pattern carries on.

Check1 + 3 + 5 + 7 = 16 = 4². The sum is n².

Examples written by our team to show the format.

Four models. The same idea, with more layers.

Compare all models

Perceptron50% rightSmall, fast and cheapThe simplest learning unit. Built for high-volume, low-latency work and on-device use.SpeedFastestReasoningDirect answersAccessAPIAbout Perceptron

Layer50% rightBalanced, the everyday workhorseA multi-layer model for most coding, writing and analysis work.SpeedFastReasoningStep by stepAccessAPIAbout Layer

Deep50% rightFlagship, most capableOur top model for hard reasoning and long, complex tasks.SpeedDeliberateReasoningExtended reasoningAccessAPIAbout Deep

By applicationMarkResearch tierResearch tier, limited accessA frontier model for specialized or sensitive work, with tighter access controls.SpeedVariesReasoningFrontierAccessBy applicationApply for access

Each figure is a small network of that model's depth (one neuron, one hidden layer, three hidden layers, four hidden layers), trained live on the same points. They stand in for the idea: the real models are large language models. All four start again from random weights in 26s.

Call it from the tools you already use.

The API speaks OpenAI's Chat Completions format, so the SDKs you have work with a new base URL. Keys, usage and billing live on the developer platform.

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.rosenblatt.alsoftworks.dev/v1",
    api_key=os.environ["ROSENBLATT_API_KEY"],
)

answer = client.chat.completions.create(
    model="rosenblatt-1",
    messages=[{"role": "user", "content": "Prove that the square root of 2 is irrational."}],
)
print(answer.choices[0].message.content)

Named after a learning machine.

Train the original perceptron

1958
Frank Rosenblatt describes the perceptron, a machine that learns from examples.
1960
The Mark I Perceptron learns simple shapes through a 20×20 grid of photocells.
1969
Minsky and Papert show what one layer can't do, like data no straight line can split.
Today
Many layers, trained deep, turn the same idea into language models.

Start building with Rosenblatt

Tell us what you're working on and we'll help you pick the right model.

Request access See API pricing