Kronos Financial Time-Series AI Model
source post: Video by cooper.simson
Video by cooper.simson
Source: instagram · Cooper Simson | Actionable AI | Agents | AI Content Saved: 20260605 Tags: instagram, ai, claude, stocks Display: Kronos Financial Time-Series AI Model — Kronos tokenizes candlestick chart data and predicts stock price movements, trained on 12B records from 45 exchanges.
TL;DR
Kronos is an AI model for financial time-series analysis that tokenizes candlestick (K-line) chart data and uses autoregressive pre-training to predict stock price movements. It was trained on 12 billion records from 45 exchanges. It addresses the lack of domain-specific foundation models for financial chart data by treating candlestick patterns as a tokenizable language, enabling price prediction without relying on general-purpose LLMs like GPT.
What the post showed
Caption: Someone built a free AI model that reads candlestick charts the way GPT reads English.
Trained on 12 billion records from 45 exchanges, outperforms every model by 93%.
Comment “STOCK” for the skill + walkthrough 📈
#ai #claude #stocks #investing #wallstreet
Key claims from transcript:
- (no transcript available)
On-screen text / OCR: Analyze / Predict Any Stock Price With Claude = S, if E m S&S i raz y — "ih fe ‘\ , 5 ~ a si tare freaking out K-line Tokenization 1 Autoregressive Pre-training of 1 =) a me ait Cae Hy Hy.%%29 I ( 3h 4 } 1 | Token ((ke + ky) bits) f i = | Vv ( Header (Header) | Coarse-grained Subtoken (ke bits) | Tokenizer Encoder \ T Query | _ Fine-grained Subtoken (Ky bits) | — WBS i Lf Cross — Ue, ‘oleae at 1 A
Extraction path:
- yt-dlp:metadata
- tesseract:ocr
Extraction warnings:
- whisper error: IndexError: tuple index out of range
What it actually is
- What: Kronos is an AI model for financial time-series analysis that tokenizes candlestick (K-line) chart data and uses autoregressive pre-training to predict stock price movements. It was trained on 12 billion records from 45 exchanges.
- Who built it / maintained by: shiyu-coder (individual researcher on GitHub)
- Status: stable
- Why it matters: It addresses the lack of domain-specific foundation models for financial chart data by treating candlestick patterns as a tokenizable language, enabling price prediction without relying on general-purpose LLMs like GPT.
- How it compares to alternatives:
- FinGPT
- BloombergGPT
- TimeGPT
- LagLlama
- PatchTST
- GitHub stars: 31,836 · License: MIT · Archived: no
Links
Kickstarter guide
Clone the repository from https://github.com/shiyu-coder/Kronos and follow the README setup instructions. The model is implemented in Python, so install dependencies via the provided requirements file. You can then load pre-trained weights and run inference on candlestick data from supported exchanges. The social media post also references using it alongside Claude for a guided walkthrough.
Retry history
- Updated: 2026-07-06
- Previous filename:
20260605-video-by-coopersimson.md - Generated filename:
20260605-video-by-coopersimson.md