# AI forecasting tools in 2026: what each one does and how to choose

> From time-series models like TimesFM to event forecasters, bot tournaments and prediction markets: the main AI forecasting tools in 2026, and how to choose one.

Oct 14, 2026 · AI forecasting · Sikt Intelligence · https://www.siktintelligence.com/blog/ai-forecasting-tools

**Short answer: it depends on what you forecast.** If you need a number over time, such as sales, demand or energy use, the leading tools are open-source models such as TimesFM, Chronos and LightGBM, or the forecasting built into planning software. If you need the probability of a real-world event, such as an election or a rate decision, the options are AI forecasting services, open benchmarks and tournaments for people who build their own, and prediction markets as the price to beat. Whatever you choose, trust only a tool with a scored track record on questions that were in the future when it forecast them.

## Two jobs, two kinds of tool

[AI forecasting](/blog/ai-forecasting-explained) covers two different jobs, and the tools rarely overlap:

- **Forecasting numbers over time:** how many units, how much load, what price. The tool learns from history, often across thousands of related series.
- **Forecasting events:** will something happen by a date? The answer is a probability, and the tool has to research a question it may never have seen before.

## Tools for forecasting numbers

- **Time-series foundation models.** Google's TimesFM and Amazon's Chronos are open-source models pretrained on large collections of time series. They can forecast a new series "zero-shot", without training on it, which makes them a quick first test on your own data ([Das et al.](https://arxiv.org/abs/2310.10688), [Ansari et al.](https://arxiv.org/abs/2403.07815)).
- **Gradient-boosted trees.** LightGBM and similar libraries are the workhorses of large retail forecasting: every top entry in the M5 competition, on 42,840 series of Walmart sales, was machine learning, mostly built on LightGBM ([Makridakis et al., 2022](https://doi.org/10.1016/j.ijforecast.2021.11.013)).
- **Classic statistical methods.** Exponential smoothing and ARIMA are not AI, but every serious comparison includes them, because on short, single series they are often as accurate.
- **Forecasting inside business software.** Many planning, CRM and cloud tools now include AI forecasting. Check what the model is and how it is tested; products change, too. Amazon Forecast, for example, closed to new customers in July 2024 ([AWS](https://docs.aws.amazon.com/forecast/latest/dg/doc-history.html)).

When AI beats the simple methods, and when it does not, is covered in [AI forecasting vs. traditional forecasting](/blog/ai-forecasting-vs-traditional-forecasting).

## Tools for forecasting events

- **General AI chatbots.** Any large language model will give you a probability if you ask. But it has no track record, often gives a different number each time, and can sound more certain than it should. Treat it as a starting point, not a forecast.
- **AI forecasting services.** Systems built specifically to forecast events: they start from a [base rate](/blog/base-rates-forecasting), research the evidence, combine several forecasts and are scored on the results. This is what an [AI superforecaster](/blog/what-is-an-ai-superforecaster) is. Examples include Sikt Intelligence (our own) and FutureSearch, which builds AI research agents that forecast for clients ([FutureSearch](https://futuresearch.ai/blog/kalshi-forecaster-case-study)).
- **Open benchmarks and tournaments.** For people who build their own forecasters, these are the fair way to measure one:
  - **Metaculus FutureEval:** bot tournaments with $175,000 in yearly prizes, including a $50,000 tournament every four months, and a human baseline of professional forecasters ([Metaculus](https://metaculus.substack.com/p/metaculus-futureeval-ai-forecasting-benchmark)).
  - **ForecastBench:** run by the Forecasting Research Institute and open to public submissions, with one leaderboard for models out of the box and one for full systems ([FRI](https://forecastingresearch.substack.com/p/put-your-ai-forecaster-to-the-test)).
  - **Prophet Arena:** a live leaderboard that scores AI models against real prediction-market prices ([Prophet Arena](https://www.prophetarena.co/)).
- **Prediction markets.** Kalshi and Polymarket are not AI tools, but their prices are the toughest everyday benchmark for any event forecast. See [how prediction markets work](/blog/how-prediction-markets-work).
- **Human superforecasters.** Good Judgment sells forecasts from the original superforecasters. They remain a strong reference point; see [AI vs. superforecasters](/blog/ai-vs-superforecasters).

The three benchmarks are compared in detail in [AI forecasting benchmarks explained](/blog/ai-forecasting-benchmarks).

## The options side by side

| Tool type | Best for | Output | Track record to check |
|---|---|---|---|
| Time-series foundation models | Quick forecasts on new series | Numbers, with ranges | Benchmarks such as GIFT-Eval, and a test on your own data |
| Gradient-boosted trees | Thousands of related series with extra data | Numbers | Competition results such as M5, and your own backtest |
| Classic statistical methods | Short, single series | Numbers, with ranges | The baseline every other tool must beat |
| General AI chatbots | A first rough estimate | A probability, unscored | Usually none |
| AI forecasting services | Real-world events | Probabilities, scored | Published, scored forecasts, misses included |
| Bot tournaments and benchmarks | Measuring your own forecaster | Rankings | The leaderboard itself |
| Prediction markets | The crowd's current view of an event | Prices | Long public history, with [known biases](/blog/favorite-longshot-bias) |

## How to choose an AI forecasting tool

Ask five questions before trusting any tool:

1. **Is there a track record on future questions?** Forecasts made before the answer existed, not backtests. A model tested on the past may simply remember what happened.
2. **Is it scored properly?** For events, the [Brier score](/blog/brier-score-explained); for numbers, scaled errors such as MASE. Hit rates and cherry-picked wins are not enough.
3. **Does it beat a simple baseline?** The base rate or the market's price for events; a seasonal naive forecast for numbers.
4. **Is it calibrated?** Do its 70% forecasts happen about 70% of the time? See [forecast calibration](/blog/forecast-calibration).
5. **Does it fit the job?** A great sales forecaster is useless for elections, and the other way round.

### Red flags

- Accuracy claims without saying what was measured, or against what.
- Only backtests, especially of language models on events from before their training cutoff.
- No misses shown.
- Promises of trading profits. Edges in prices shrink as soon as they are widely used; see [can AI predict the stock market?](/blog/can-ai-predict-the-stock-market)

## Where Sikt Intelligence fits

Sikt Intelligence is an AI forecasting tool for real-world events: rate decisions, elections, geopolitical flashpoints and technology milestones. Each forecast is a probability shown next to the market's price, and each one is graded when the answer arrives, misses included. You can follow it live on the [2026 midterms](/midterms) and the [odds pages](/odds). Sikt is in research; [join the waitlist](/#access) for early access. Nothing here is financial or investment advice.

## Key takeaways

- Choose by the job: numbers over time and real-world events need different tools.
- For numbers, test open-source models such as TimesFM, Chronos and LightGBM against a simple statistical baseline on your own data.
- For events, look for AI forecasting services with a public, scored track record, and compare them with prediction-market prices.
- Metaculus FutureEval, ForecastBench and Prophet Arena are the fair ways to measure a forecaster you build yourself.
- Trust only forecasts made before the answer existed, scored with proper rules, misses included.

## FAQ

### What is the best AI forecasting tool?

There is no single best tool. For numbers over time, foundation models such as TimesFM and Chronos and gradient-boosted trees such as LightGBM lead public benchmarks and competitions. For real-world events, the best choice is an AI forecasting service with a public, scored track record that holds up against prediction-market prices.

### Can I use ChatGPT for forecasting?

You can ask any AI chatbot for a probability, but the answer has no track record, can change each time you ask and is often overconfident. Purpose-built forecasting systems research the question, combine several forecasts and are scored on the results.

### Are there free AI forecasting tools?

Yes. TimesFM, Chronos and LightGBM are open source, and Metaculus and ForecastBench publish their questions and leaderboards openly. Prediction-market prices are also free to view.

### How do I know if an AI forecast is accurate?

Look at many of its forecasts, made before the answers were known, scored with a proper rule such as the Brier score, and compared with a simple baseline or the market's price. One correct call proves very little.

## Sources

- Das et al. (2024): [A decoder-only foundation model for time-series forecasting](https://arxiv.org/abs/2310.10688) (TimesFM)
- Ansari et al. (2024): [Chronos: Learning the Language of Time Series](https://arxiv.org/abs/2403.07815)
- Makridakis, Spiliotis and Assimakopoulos (2022): [M5 accuracy competition: Results, findings, and conclusions](https://doi.org/10.1016/j.ijforecast.2021.11.013), *International Journal of Forecasting*
- AWS: [Amazon Forecast document history](https://docs.aws.amazon.com/forecast/latest/dg/doc-history.html)
- Metaculus: [FutureEval, Metaculus's AI forecasting benchmark](https://metaculus.substack.com/p/metaculus-futureeval-ai-forecasting-benchmark)
- Forecasting Research Institute: [Put your AI forecaster to the test on ForecastBench](https://forecastingresearch.substack.com/p/put-your-ai-forecaster-to-the-test)
- Prophet Arena: [AI forecasting benchmark on real prediction markets](https://www.prophetarena.co/)
- FutureSearch: [Kalshi forecaster case study](https://futuresearch.ai/blog/kalshi-forecaster-case-study)