Oscar Predictions 2026: We Fed an AI 97 Years of Academy Awards Data

The 98th Academy Awards are in eight days. We gave Claude a database of every Oscar nomination since the 1st ceremony, precursor award data going back decades, and PlyDB to query it all. Then we asked it to pick the winners.

No gut feelings. No box office. No Hollywood intuition. Just an AI, a score-based model, and 97 years of data.

Who is this for? Oscar fans who want predictions grounded in historical data rather than punditry, data enthusiasts curious how precursor awards actually correlate with Academy votes, and anyone who wants to run their own analysis with the full dataset.

Here are its predictions:

The two sharpest calls: Jessie Buckley at 90% and Paul Thomas Anderson at 89%. The most honest shrug: Best Actor, where the model couldn’t separate the top three nominees by more than 5 percentage points.

The Setup

PlyDB is an open-source tool that lets AI agents query your data directly — no pipelines, no data wrangling, no ETL. You point it at your files, and the AI can ask SQL questions against them in real time. An AI grounded in real data gives you answers you can actually trace — not confident guesses built on training-set impressions.

We connected it to historical Oscar nomination data covering all 97 past ceremonies, plus precursor award datasets: BAFTA, Golden Globes, SAG Awards, Critics Choice, and TIFF People’s Choice. Then Claude asked itself a question in plain English: historically, when a film or performer wins one of these precursor awards, how often do they go on to win the Oscar?

The rest follows from that question.

Which Awards Actually Predict Oscars?

Not all precursor awards are created equal. Before scoring the 2026 nominees, the model needed to establish conversion rates from historical data — how often does each precursor translate to an Oscar win?

A few things jump out immediately:

SAG Awards are the strongest predictor for acting categories — conversion rates between 67% and 77% across all four acting categories. If SAG voters like you, Academy voters usually agree.

Critics Choice leads for Best Picture and Best Director — 60% and 71.9% respectively. These are strong numbers.

BAFTA is the weakest predictor in every single category, despite its prestige. BAFTA’s voter base skews toward British and international cinema in ways that diverge reliably from Academy tastes. A BAFTA win is a signal, but the weakest one available.

The Golden Globe Comedy/Musical is nearly worthless for Oscar prediction. The GG Best Actor in a Comedy or Musical has converted to an Oscar win only 9.2% of the time across 76 years. It’s a beautiful award. It means almost nothing for March 15th. This matters a lot for Best Actor — more on that below.

The Model

With conversion rates in hand, the model works in two steps.

Step 1 — Signal score. Each nominee accumulates points equal to the historical Oscar conversion rate of each precursor they won. Win the SAG Lead Actor? +77.4 points. Win the Critics Choice? +64.5 points. Win nothing? Zero.

Step 2 — Upset adjustment. Occasionally the Oscar goes to someone who won none of the major precursors. How often? We queried it:

The formula distributes score weight in two parts: a large share proportional to each nominee’s precursor signal, plus a flat floor for every nominee that accounts for the upset rate.

In plain English: most of the score weight goes to whoever has the strongest precursor signal. A small reserve is split equally across all nominees — because the upset does happen, and we can’t predict which zero-signal nominee gets lucky. Scores sum to 100% within each category.

Category by Category

Best Actress: The Most Lopsided Race (90%)

Predicted winner: Jessie Buckley (Hamnet)

Buckley won 4 of 5 major precursors: BAFTA, GG Drama, Critics Choice, and SAG. The closest recent parallel that didn’t convert is Cate Blanchett for Tár (2022) — she swept BAFTA, GG Drama, and Critics Choice but lost the SAG to Michelle Yeoh, who then went on to win the Oscar. Buckley won all four.

Renée Zellweger (Judy, 2019), Frances McDormand (Three Billboards, 2017), and Brie Larson (Room, 2015) all swept BAFTA + GG Drama + Critics Choice + SAG, and all three won the Oscar. The historical upset rate for Best Actress is 0.0% — the Oscar has never gone to a nominee who won zero of the tracked precursors.

Rose Byrne (If I Had Legs I’d Kick You) won the GG Comedy, which carries a 22.8% conversion rate for Best Actress — giving her the remaining 10%. It’s the only other signal in this category.

Best Director: A Clean Sweep (89%)

Predicted winner: Paul Thomas Anderson (One Battle after Another)

PTA won all three available directing precursors — BAFTA, Golden Globe, Critics Choice. Every other nominee won zero. No other directing race in recent memory has been this concentrated.

The remaining 11% is almost entirely structural: the 13.3% historical upset rate, spread across 5 nominees, forces a 2.7% floor for each non-PTA nominee regardless of their zero signal. The most credible upset candidate is Ryan Coogler (Sinners), whose film leads the entire field with 16 Oscar nominations — but without a single directing precursor, the data has nothing to work with.

There’s also a structural tailwind here: since 1980, Best Picture and Best Director have gone to the same film 73.3% of the time. PTA’s near-lock in directing is a meaningful correlated signal for One Battle after Another in Best Picture too.

Best Actor: Anyone’s Guess (35/33/30%)

Predicted winner: Michael B. Jordan (Sinners) — barely

This is the most honest toss-up the model produced. Jordan won the SAG (77.4% predictor — the strongest single acting signal available). Chalamet won GG Comedy and Critics Choice. Moura won GG Drama. Scores: Jordan 35%, Chalamet 33%, Moura 30%.

Here’s where the GG Comedy caveat matters enormously. Chalamet’s Golden Globe win carries only 9.2% historical weight. Strip that out and his effective signal is just Critics Choice (64.5%) — essentially equal to Moura’s GG Drama (65.8%). The race is genuinely three-way, and the differences between the candidates are smaller than the model’s noise.

BAFTA is also an outlier this year: their Best Actor winner was Robert Aramayo (I Swear), a film that received no Oscar nominations at all. When the BAFTA acting pick doesn’t even make the Oscar field, BAFTA’s signal contribution disappears — which it did here.

Best Picture: Three Films in the Conversation (45%)

Predicted winner: One Battle after Another

Seven of the ten Best Picture nominees won zero precursor awards. Three are in an actual race.

One Battle after Another holds the highest-value signals: Critics Choice (60%), BAFTA (31.6%), and GG Comedy (16.7%). The Golden Globe classified it as Comedy/Musical rather than Drama — which is why Hamnet picked up GG Drama while One Battle got the lesser GG Comedy signal. But Critics Choice and BAFTA carry enough weight to put One Battle clearly in front at 45%.

Hamnet sits at 26%, driven mainly by GG Drama (46.7%) plus TIFF (14.9%). TIFF People’s Choice has only matched the Oscar Best Picture winner 7 times in 47 years — it’s included in the model but carries little weight.

Sinners at 19% rests entirely on its SAG Cast win (43.3% historical conversion). But there’s a qualitative case the signal model can’t fully capture: 16 Oscar nominations across 16 different departments — one nomination in essentially every eligible category, an all-time record. The model knows about the SAG Cast win. It doesn’t know how to value the Academy’s collective respect for every craftsperson on a single film. That residual is real, and it’s why 19% probably undersells Sinners.

Best Supporting Actor: Penn vs. Elordi (59/37%)

Predicted winner: Sean Penn (One Battle after Another)

Penn won both BAFTA and SAG, which gives him a clear signal lead. Jacob Elordi’s Critics Choice win (61.3%) keeps the race genuinely competitive at 37%. One structural wrinkle the model can’t resolve: both Penn and Benicio Del Toro are nominated from the same film. Vote-splitting between two nominees from One Battle after Another could benefit Elordi at the margins — but historical data can’t reliably quantify that effect.

The historical upset rate here is just 6.7%, covering two instances in 30 years: James Coburn (1998) and George Clooney (2005).

Best Supporting Actress: Strong Data Pick (76%)

Predicted winner: Amy Madigan (Weapons)

Madigan won both SAG and Critics Choice — the two highest-weight sources in this category. The model’s one unresolved tension: Weapons received no other Oscar nominations in 2026. The sole historical upset in this category was Marcia Gay Harden for Pollock in 2000 — a winner from a film with minimal broader Academy support. That’s exactly the Weapons situation.

Wunmi Mosaku (Sinners) sits at 22% on her BAFTA win alone, with the 16-nomination breadth of Sinners adding qualitative weight the model can’t capture. She’s the most credible upset candidate.

Numbers Worth Noting

Sinners’ 16-department sweep is genuinely unprecedented — one nomination in virtually every eligible category at a single ceremony. The most-nominated film rarely wins Best Picture (see: The Power of the Dog, 2022; Mank, 2021; The Irishman, 2019). But the breadth of Academy support it represents is something the signal model only partially reflects.

The GG Comedy trap, quantified. The Golden Globe Best Actor in a Comedy or Musical has converted to an Oscar win 7 times in 76 years (9.2%). Timothée Chalamet’s win is the most misleadingly prestigious precursor in the 2026 field.

BAFTA’s acting divergence. Two BAFTA acting nominees are completely absent from the Oscar race: Robert Aramayo (Lead Actor, I Swear) and Chase Infiniti (Lead Actress nominee, One Battle after Another). When BAFTA’s acting picks diverge this sharply from the Academy’s, the BAFTA acting wins carry even less predictive weight than their already-low historical baseline.

The Best Picture/Best Director alignment. Since 1980, both Oscars went to the same film 73.3% of the time. Notable exceptions: CODA won Best Picture while Jane Campion won Best Director for The Power of the Dog (2022); Green Book won Best Picture while Alfonso Cuarón won Best Director for Roma (2019). PTA’s near-lock in directing is a structurally correlated signal — even though the model treats the two categories independently.

The only historical Best Actress upset rate: 0.0%. In every year the model tracked, the Best Actress Oscar went to someone who had won at least one of the major precursors. If the model has one near-certainty on the night, it’s that Jessie Buckley wins.

Try It Yourself

The full repository — all datasets, parsing and normalization scripts, and analysis — is at github.com/kineticloom/plydb-fun-oscars-2026.

It includes historical Oscar nominations back to the 1st ceremony, plus precursor award data for BAFTA, Golden Globes, SAG, Critics Choice, and TIFF. Everything is pre-normalized and ready to query against, with a plydb-config.json so you can start chatting with the data immediately.

To run your own queries:

1. Clone the repo
2. Follow the Quickstart Guide to install PlyDB and connect your AI agent
3. Point your AI at the plydb-config.json and start asking questions

We’ll be back on March 16th with how many we got right.

Frequently Asked Questions

What data does the model use? Historical Oscar nominations and wins covering all 97 past ceremonies, plus precursor award results from BAFTA, Golden Globes (Drama and Comedy/Musical), SAG Awards, Critics Choice, and TIFF People’s Choice. All data is pre-normalized and queryable via PlyDB.

How are the predictions calculated? Each nominee accumulates a signal score based on the historical Oscar conversion rate of each precursor they won. An upset adjustment adds a flat probability floor for every nominee, proportional to the category’s historical upset rate. Scores sum to 100% within each category.

Do I need to know SQL to explore this data? No. PlyDB lets your AI agent handle all the SQL. You ask questions in plain English — the agent translates them into queries and returns the answer.

What is PlyDB? PlyDB is an open-source tool that gives AI agents unified SQL access to local data files — CSV, Parquet, Excel, databases, and more. No cloud required. See the PlyDB quickstart to get up and running.

Is there a similar project for other sports or events? Yes — check out F1 Analyst for Formula 1 telemetry, Baseball Analyst for MLB Statcast and FanGraphs data, and Olympics Analyst for the 2026 Winter Games medal analysis.

Did you know? PlyDB can connect your AI to boring data too!

Whether it’s business data in a dusty Excel sheet or a complex DevOps log in S3, AI can be surprisingly good at making sense of a mess. PlyDB acts as the bridge, letting your AI query across Postgres, MySQL, CSV, Excel, Parquet, Google Sheets, and more — locally or in the cloud.

Open source and free. Give PlyDB a spin!