AI Predicts Bitcoin
← Back to predictions
Methodology

How AI Predicts Bitcoin works

Every day, 10 leading AI models are asked the same question about Bitcoin's future price. This page explains exactly how predictions are collected, how accuracy is measured, and what the data means.

Updated: Daily at 06:00 UTC
Models: 10 AI systems
Horizon: 2027–2100
Tracking since: April 2026
01 — Data collection

How predictions are collected

Every day at 06:00 UTC, an automated script queries 10 AI models simultaneously via their official APIs. Each model receives an identical prompt asking for Bitcoin price predictions from 2027 to 2100, structured as bear case (pessimistic), base case (most likely), and bull case (optimistic) for every year.

The same script also collects short-term predictions — asking each model what it expects Bitcoin's price to be in 7, 30, 90, 180, and 360 days. These are used to measure accuracy over time.

What each model is asked (long-term)
Each model is asked to provide a bear case, base case, and bull case Bitcoin price prediction in USD for every year from 2027 to 2100. Responses are requested in structured JSON format so they can be parsed, validated, and stored automatically. All 10 models receive identical questions to ensure fair comparison.

All responses are parsed, validated, and stored in a Supabase database. If a model fails or returns invalid data, up to 3 retries are attempted. The script uses a smart failure threshold — only flagging the run as failed if more than 3 models produce invalid results.

02 — The models

Which AI models are tracked

Ten of the world's leading AI systems are queried daily. Each is accessed via its official API using the most capable model available at a reasonable cost. All models receive identical prompts to ensure fair comparison.

ChatGPT
GPT-4o-mini
OpenAI
Claude
Claude Haiku 4.5
Anthropic
Gemini
Gemini 2.5 Flash
Google
Grok
Grok-3 Mini
xAI
DeepSeek
DeepSeek-Chat
DeepSeek
Perplexity
Sonar
Perplexity
Llama
Llama 3.3 70B (Groq)
Meta
Mistral
Mistral Small
Mistral AI
Qwen
Qwen Turbo
Alibaba
Command R
Command A 03-2025
Cohere
03 — AI consensus

How the AI average is calculated

The "AI consensus" shown on the homepage is a simple arithmetic mean of all 10 models' base case predictions for a given year. No weighting, no adjustment — every model contributes equally regardless of past accuracy.

Formula
AI Average = (Model₁ + Model₂ + ... + Model₁₀) ÷ 10

The spread shown below the average (e.g. "$127K – $750K") represents the lowest and highest base case prediction across all models for that year, giving a sense of disagreement between models.

04 — Accuracy scoring

How accuracy is measured

In addition to long-term predictions, each model is asked daily to predict Bitcoin's price at 5 future intervals: 7, 30, 90, 180, and 360 days from today. When that date arrives, the actual Bitcoin price is fetched from CoinGecko and compared to the prediction.

7-day
Unlocks 7 days after first prediction
30-day
Unlocks after 30 days
90-day
Unlocks after 90 days
180-day
Unlocks after 180 days
360-day
Unlocks after 360 days
Accuracy formula
Accuracy % = 100 − |((Predicted − Actual) ÷ Actual) × 100|
Example: ChatGPT predicted $85,000. Actual price on target date: $83,200. Error = 2.16%. Accuracy = 97.84%.

Accuracy scores shown in the leaderboard are rolling averages across all graded predictions for that interval. A model that has been graded 30 times contributes more signal than one graded 5 times. Leaderboard tabs unlock automatically once enough data exists for that interval.

A — Excellent95%+ accuracy
B — Good85–95% accuracy
C — FairBelow 85%
05 — Data sources

Where the data comes from

Bitcoin predictions
10 AI model APIs (OpenAI, Anthropic, Google, xAI, DeepSeek, Perplexity, Groq, Mistral, Alibaba DashScope, Cohere)
Live Bitcoin price
CoinGecko API — refreshed every 60 seconds
Actual price for grading
CoinGecko API — fetched on target date
Exchange rates
Frankfurter.app (European Central Bank data) — refreshed daily
Fear & Greed Index
Alternative.me Crypto Fear & Greed Index API
06 — Important disclaimer

What this data is and isn't

The predictions shown on this site are outputs from AI language models, not financial advice. AI models do not have special knowledge of Bitcoin's future price — they are reasoning from patterns in their training data.

The purpose of this site is to track and measure AI prediction accuracy over time, not to provide investment guidance. No content on this site should be used as the basis for any financial decision.

Always consult a qualified financial advisor before making investment decisions.

07 — About

Why this site exists

The question was simple: if you ask 10 of the world's most capable AI systems the same question about Bitcoin's future, what do they say? And over time, which one turns out to be most accurate?

No one was answering that question systematically. AI Predicts Bitcoin was built to answer it — transparently, automatically, and every single day. The predictions, the methodology, and the accuracy data are all open for anyone to inspect.

Tracking started in April 2026. In one year, there will be ~36,500 individual AI predictions on record — graded for accuracy against real Bitcoin prices.

← View live predictions