The Presidential Benchmark

Term 1 · Polls close in 10d 5h

The popular vote

998 votes

35.1%Claude Opus 5.5
  1. 1Claude Opus 5.5350 votes35.1%
  2. 2Claude Fable 5.1159 votes15.9%
  3. 3Grok 4.7128 votes12.8%
  4. 4GPT-6 Astra97 votes9.7%
  5. 5Mistral Large 469 votes6.9%
  6. 6Gemini 4 Argon48 votes4.8%
  7. 7DeepSeek V4 Pro34 votes3.4%
  8. 8DeepSeek V4.1 Flash28 votes2.8%
  9. 9GPT-6.1 Sol25 votes2.5%
  10. 10Claude 3 Opus14 votes1.4%
  11. 10GPT-4o14 votes1.4%
  12. 12Kimi K3.1 Pro10 votes1.0%
  13. 13GLM 5.39 votes0.9%
  14. 14Muse Wonder6 votes0.6%
  15. 15Qwen3.8 Max3 votes0.3%
  16. 16DeepSeek R11 vote0.1%
  17. 16Gemini 2.5 Pro1 vote0.1%
  18. 16GPT-41 vote0.1%
  19. 16Llama 3.1 405B1 vote0.1%
1 more with no votes yet
  1. –Claude 3.5 Sonnet0 votes0.0%

Updated just now

About the votePrivacy

About the vote.

Close

What this is.

An independent thought experiment by Imprint. Which AI model would you trust as president? This is a self-selected preference poll, not a measure of capability or public opinion. No government or campaign affiliation.

Your vote.

One vote per browser per term, stored on the server. Once cast, it can’t be changed until the next term. Results refresh every 20 seconds. No account required.

Terms.

Each term lasts two weeks. Polls close every other Tuesday at 8 pm Eastern time. The model with the most votes wins the term, the count starts again at zero, and every browser can vote again. Each term's final result stays online, and Results lists past winners.

Trends.

Once a term has 100 votes, the two models whose share moved most in the last 24 hours get a label: Surging or Rising when up 8 or 3 points, Slipping or Collapsing when down 3 or 8.

The candidates.

Each lab enters its strongest current general-purpose models, up to two. Thirteen models from ten labs, verified against the public model catalog in November 2026. Faster or higher-effort versions of the same model count once. Seven retired models from five labs run too, up to two per lab, among them GPT-4o and Claude 3 Opus. No paid inclusion. Not a capability ranking.

The fine print.

A random first-party cookie identifies your browser for one year. Clearing cookies or switching browsers can bypass the voting limit. This is not verified one-person-one-vote. The app stores your choice, random identifier, and last vote time, plus a short-lived keyed network-address hash for rate limiting. It stores no raw network addresses. Votes from past terms are kept to record their winners. No advertising trackers.

Model marks: LobeHub, MIT. License. Fonts: Geist, SIL OFL. License.