Flow - 4x faster than typing
Mini toolstalkie 1930 13b base
By talkie
talkie is a 13B historical language model designed to simulate a conversation partner from 1930 rather than a modern assistant. Its creators say the base model was trained on 260B tokens of pre-1931 English text drawn from sources such as books, newspapers, scientific journals, patents, and case law, then post-trained into a conversational checkpoint without using modern chat transcripts or standard instruction-tuning data. It is positioned as a research vehicle for studying vintage language models, historical knowledge cutoffs, and data-contamination-free evaluation.
New
Text
Overview
talkie is a 13B vintage language model trained only on pre-1931 English text. It is built to act as a historically grounded conversation partner, with a 1930 knowledge cutoff, and is positioned as the largest vintage language model its creators are aware of.
About talkie
Independent AI research lab developing "vintage language models" — LLMs trained exclusively on historical text. Their flagship model, talkie-1930-13b, is a 13B-parameter model trained on 260B tokens of pre-1931 English text (books, newspapers, patents, case law), released under Apache 2.0. The project explores contamination-free generalization, temporal forecasting, and the impact of data diversity on model behavior. Founded by Nick Levine, David Duvenaud, and Alec Radford.
View Company Profile
Last updated: July 7, 2026
