FutureSearch Forecaster

Forecast outcomes, timelines, valuations

Forecasting agents research sets of questions to produce probabilities, dates, or numbers about the future. Accuracy backed by our forecasting evals.

Try it for freeGive to your AI

▶ 2-min demo video coming soon

Diagram showing companies like Anthropic and OpenAI with forecasted IPO valuations: $630B, $1.0T, $350B
$0.09 to $0.20default · $1.20 high effort
Binary, numeric, and dateforecast types
Chart showing forecasted seed valuations for AI researchers like Noam Brown, Geoffrey Hinton, and Alec Radford

AI Researcher Startup Valuations

Forecast seed valuations for 116 AI researchers if they left to found a startup.

116 researchers forecasted

Results table showing Polymarket questions sorted by annualized ROI

Polymarket Profit Screening

Forecast 100 prediction market questions to find mispriced contracts, then sort by expected ROI.

31 profitable questions identified

Chart showing forecasted first-day market cap for Anthropic and OpenAI IPOs

Anthropic & OpenAI IPO Forecasts

Forecast IPO timelines and first-day valuations for the two leading AI labs.

Date + valuation + reasoning

Give your AI a team of forecasters

Claude Code

claude mcp add futuresearch --scope project --transport http https://mcp.futuresearch.ai/mcp

Then ask Claude to forecast your data.

Python SDK

pip install futuresearch

from futuresearch.ops import forecast
result = await forecast(
  task="Will this person leave
    to found an AI startup
    by end of 2027?",
  input=researchers_df,
)

Pricing

Start with $20 in free credits. No credit card required. Costs vary based on the complexity of research needed to produce each forecast.

Case studyRowsForecast typeCost
AI researcher startup valuations116Numeric (percentiles)$9
Polymarket profit screening55Binary probabilities$5
Anthropic & OpenAI IPO forecasts2Date + valuation$1

Why costs vary

You only pay for the intelligence each row needs. Simple binary questions cost less; numeric forecasts with full percentile distributions require more research. Default effort resolves automatically: HIGH for a single forecast, LOW for many. Accuracy is independently measured on BTF-2, described in the Strategic Reasoning paper, with the dataset on Hugging Face.

Date mode returns a full distribution over timing as {output_field}_p10 … _p90 YYYY-MM-DD strings, with the literal string "never" reserved for percentiles in the indefinite future.

Resources

Guides

SDK Docs

Related

Start forecasting your data

Start with $20 free credit. No credit card required.

Try it for freeSee pricing