My 2026 Hugo Winner Picks

Here’s who my model is picking to win each of the big five 2026 Hugo Awards, and the probability I’d put on each win. Finalists were announced April 21. Ceremony is August 30 at LAcon V in Anaheim.

The picks

You’ll note that none of these predictions are that much above the baseline 17% probability. That’s indicative of the fact that predictions are hard, especially about the future. “In My Country” has the clearest margin over the rest of its field. The other four top picks are favorites, with a lot of probability still spread across the other finalists in each category.

How I’m doing this

I’m building this script because my previous Hugo predictions have not been particularly good. In the past my process amounted to: look up Mr Philip’s Library’s predictions, then adjust a bit based on whether I had personally read the book and whether I’d be ranking it highly on my own Hugo ballot.¹ Everything I’ve read about making good predictions says it’s important to have a model rather than going with your gut. So here we are.

I put the model together this year and backtested it against a handful of prior Hugo years. Some of the signals I’d want are hard to reconstruct historically, so the backtest is less clean than I’d like. The model takes the finalist list and produces ranked win probabilities, which I blend with current Manifold market prices on a weighted basis so that thin markets carry less weight than the model.

A couple of patterns

The Nebula novelette shortlist has been a weak negative signal for winning the Hugo novelette over recent years. In other short-fiction categories the Nebula signal stays positive. Novelette is the exception. The sample is small, so this could just be noise. It’s weird enough that I’m planning to push the dataset further back to see if it holds.

The Manifold prediction markets for past Hugo winners have been wrong more often than right. On average, the day before the ceremony, these markets have put less probability on the eventual winner than a completely uninformed flat prior would. I also tried posting a Hugo Best Novel question on Metaculus in 2024 and got a total of four forecasters, including me. Both the small community and my own forecast put the eventual winner below uniform. Thin markets on obscure topics don’t aggregate information, which matches what Tetlock’s work on forecasting would predict. Crowd wisdom needs enough crowd.

After August 30

I’ll post a followup with how the picks actually did, and whatever that does to my confidence in the methodology.

Caveats: these are subjective posteriors fit on a small sample. Expect wide error bars. Most likely a mix of hits and misses.