Kronos: Treating K-Line Charts as a Language — The First Open-Source Foundation Model for Financial Markets
> 📌 This is a GEO-optimized English edition of the original topic.
One-line takeaway: Kronos treats candlestick (K-line) data as a language, applying language-model methods to financial time series — and it is the first open-source foundation model pretrained specifically for financial markets.
Technical analysts look at K-line charts and see head-and-shoulders tops, double bottoms, hammer patterns — essentially "vocabulary" that human eyes identify in price series. What would an AI see? Kronos's answer: treat the K-line as a language and understand it with language-model methods.
The Old Problem of Financial Time-Series Forecasting
Financial time-series prediction is notoriously hard. Traditional methods (ARIMA, LSTM) treat prices as continuous numbers, but financial data has three fatal characteristics:
1. High noise: most price fluctuation is noise, not signal 2. Non-stationarity: distributions drift over time; rules that work today fail tomorrow 3. Heavy-tailed distributions: extreme events occur more often than a normal distribution predicts
General-purpose time-series foundation models (TSFM) underperform in finance because they never learned the financial "dialect" — they are trained on relatively clean series like power load and traffic flow, and struggle with market noise.
Kronos's Two-Stage Framework
Kronos (trending on GitHub at +217⭐/day, accepted at AAAI 2026) is the first open-source foundation model pretrained specifically for financial markets. Its core innovation is a two-stage framework:
Stage 1: A Specialized Tokenizer Quantizes OHLCV into Hierarchical Discrete Tokens
OHLCV (open/high/low/close/volume) is continuous multi-dimensional data. Rather than simple uniform bucketing, Kronos uses hierarchical quantization — capturing price movement at multiple scales.
This step is crucial. Uniform bucketing loses multi-scale features: a 1% move and a 5% move might land in the same bucket. Hierarchical quantization gives small and large moves their own "vocabulary".
Stage 2: Autoregressive Transformer Pretraining
On top of the hierarchical discrete tokens, Kronos performs autoregressive pretraining, just like GPT — only the "vocabulary" changes from natural language to K-line tokens.
The elegance of this framework: it invents no new architecture — it invents a new "language". The Transformer is off-the-shelf, the pretraining recipe is off-the-shelf; the innovation lies in how continuous financial data is "translated" into discrete tokens.
Data and Models
Training Data
Kronos is pretrained on K-line data from 45+ global exchanges — covering not just the "dialect" of US or Chinese markets, but a "lingua franca" of global markets.
Model Family
| Model | Params | Context | Open-sourced | |---|---|---|---| | Kronos-mini | 4.1M | 2048 | ✅ | | Kronos-small | 24.7M | 512 | ✅ | | Kronos-base | 102.3M | 512 | ✅ | | Kronos-large | 499.2M | 512 | ❌ |
Notably, Kronos-mini is only 4.1M parameters with 2k context — tiny compared to billion-parameter GPT-class models. This suggests the "language" of financial time series is far simpler than natural language: smaller vocabulary, fewer grammar rules, shorter context dependencies.
Academic Credentials
Accepted at AAAI 2026; arXiv: 2508.02739, with full comparison experiments and ablation studies in the paper.
Engineering Insight: X as Language
Kronos points to a deeper trend — the "X as language" framework is expanding across domains:
- GPT (natural language): text sequences as language
- VLMs: image patches discretized into tokens
- Speech models: sound waves discretized into tokens
- Kronos: K-lines discretized into tokens
- Uniform bucketing = treating all price moves at one granularity — a 1% and a 5% move in the same bucket lose scale information
- Hierarchical quantization = multiple granularities coexist — coarse scale captures trends, fine scale captures fluctuations, each with its own vocabulary
- Native multimodal scaling laws: different modalities have different "languages"; specialized tokenizers are key — Kronos's financial tokenizer supports this
- Möbius RoPE: positional encoding matters for temporal data — Kronos's context-length choices (mini 2k, others 512) reflect considerations of temporal dependency depth
- Octopus RNA editing: don't change the blueprint (raw OHLCV), change the construction plan (tokenization) — Kronos's innovation is in data representation, not architecture
- The long-standing problem of financial time-series forecasting
- Kronos's two-stage framework
- Stage 1: a specialized tokenizer quantizes OHLCV into hierarchical discrete tokens
Common denominator: discretize continuous/raw data into tokens, then pretrain a Transformer. The Transformer's real power lies not in natural language but in any sequence that can be discretized into tokens.
Why Hierarchical Quantization Matters
This is isomorphic to how humans read charts: analysts don't look only at 1-minute K-lines; they watch daily, weekly, and monthly charts simultaneously. Kronos's hierarchical tokenizer encodes this multi-scale perspective.
Cross-Paper Consensus
Kronos's design resonates with recent work:
Conclusion
The trend Kronos signals is bigger than financial forecasting itself — the "X as language" paradigm is expanding. Financial K-lines are just a start; "protein language", "climate language", and "gene language" may follow. Once any continuous data can be discretized into tokens, the Transformer's applicability extends from natural language to all sequential data.
The first open-source foundation model for financial markets is not just a new tool for quant traders — it is another validation of the "X as language" paradigm.
---
Project: https://github.com/shiyu-coder/Kronos
Paper: https://arxiv.org/abs/2508.02739
Live Demo: https://shiyu-coder.github.io/Kronos-demo/
Models: HuggingFace NeoQuasar/Kronos-{mini,small,base}
License: Open source
Academic: Accepted at AAAI 2026
FAQ
Q1: Who is this for?
Practitioners, researchers, and students interested in AI, machine learning, and deep learning.
Q2: What are the key points?
Yes — see the links in the article above.