QANTA 2026: Efficient, Incremental Multimodal Question Answering
QANTA 2026 is the world’s first multimodal Quizbowl computer competition — questions combine text and images, testing whether AI systems and human players can reason across modalities together.
Building on the human–AI collaboration format introduced in 2025, QANTA 2026 raises the stakes: clues may include photographs, diagrams, artworks, maps, or scientific figures. Expert humans read images naturally — can AI do the same? How much can computers help humans answer questions?
Tournament Stats
Stats and standings from the QANTA 2026 tournament.
How to Participate
There are three ways to join QANTA 2026:
Build a Multimodal AI Teammate
Design a system that can interpret both text clues and images to answer quiz bowl questions. AI must handle visual reasoning, OCR, diagram interpretation, and cross-modal fusion — alongside the natural language understanding required for text-only questions.
Play as a Human
Bring your trivia knowledge and visual reasoning skills. Team up with an AI agent to tackle multimodal questions — and help us understand how humans and AI complement each other across modalities.
Write Multimodal Questions
Author pyramid-style questions that incorporate images alongside text clues. Receive $5 per accepted question. Questions should be adversarial for AI while remaining solvable by expert humans using the image and text together.
Results
The post-event materials highlight a few winners and takeaways from QANTA 2026:
- Human team winners included Groove and CoPR.
- Heavyweight champions and runners-up were Moksh Jain and Ronit Singh.
- Middleweight champions and runners-up were George Li and Nirjhar Das.
- Best question went to Jamie Carlson.
- Most adversarial question and packet went to Chauncey Lo.
- Best packet went to Stephen Pachucki.
The bigger lesson from the deck is that multimodal systems are strong but still brittle on surprising visual clues, explanation quality, calibration, and adversarial robustness.
Supported By
This year’s competition was supported by TRAILS, NSF, and Adobe.
NSF
Competition structure
Live matches use two kinds of questions, standard in quiz bowl:
-
Tossups are read clue by clue; anyone on either side can interrupt when they think they know the answer. That format rewards knowing when to commit and when to hold back — when to buzz with an answer versus abstaining until more evidence appears. Humans and computer systems both face the same pressure to calibrate confidence under incomplete clues.
-
Bonuses are team questions, usually in several parts on a shared theme. They reward depth of knowledge and the ability to weigh explanations and rationales (including from AI teammates) before the captain gives the team’s final answer.
The team that answers a tossup correctly earns the next bonus. For full match flow, interrupt rules, and scoring, see the rules hub and the human and computer rulebooks.
Join the community on Discord.
Rules and Scoring
The overview explains how tossups and bonuses chain together in a match. Human rules cover table scoring, muting, and live procedures; computer rules cover system outputs and how leaderboards score submissions.
Contact
Questions about QANTA 2026? Email qanta@googlegroups.com.