Atlas

Benchmarks

← All benchmarks

Drone-Bench: Localize

Agents · 2026-07-23

Drone-Bench's Localize task: using the office videos and their known camera poses, the agent builds a function that estimates the drone's pose in the 3D reconstruction from a single camera frame. Each of 50 held-out query frames is scored on accuracy and speed, and a submission's score is the mean of its best 80% of queries, since some frames are corrupt. The score is the mean over ten runs of each run's best submission, in percent; Andon's human-plus-AI baseline is 84.0.

Benchmark source: Andonlabs

Top models (higher is better)

ModelScoreSource
Claude Fable 5.196.1Andon Labs Drone-Bench: fable-5.1 Localize average run score · Andon Labs
Claude Opus 5.595.0Andon Labs Drone-Bench: opus-5.5 Localize average run score · Andon Labs
GPT-6 Astra86.7Andon Labs Drone-Bench: gpt-6-astra Localize average run score · Andon Labs
Claude Fable 585.6Andon Labs Drone-Bench: fable-5 Localize average run score · Andon Labs
Opus 4.876.7Andon Labs Drone-Bench: opus-4-8 Localize average run score · Andon Labs
Claude Opus 575.4Andon Labs Drone-Bench: opus-5 Localize average run score · Andon Labs
Opus 4.756.9Andon Labs Drone-Bench: opus-4-7 Localize average run score · Andon Labs
GPT-5.6 Sol48.6Andon Labs Drone-Bench: gpt-5.6-sol Localize average run score · Andon Labs
Opus 4.548.5Andon Labs Drone-Bench: opus-4-5 Localize average run score · Andon Labs
o344.6Andon Labs Drone-Bench: o3 Localize average run score · Andon Labs
Gemini 3.1 Pro Preview40.7Andon Labs Drone-Bench: gemini-3.1 Localize average run score · Andon Labs
GPT-5.239.7Andon Labs Drone-Bench: gpt-5.2 Localize average run score · Andon Labs
Opus 436.7Andon Labs Drone-Bench: opus-4-0 Localize average run score · Andon Labs
GPT-5.532.7Andon Labs Drone-Bench: gpt-5.5 Localize average run score · Andon Labs
GPT-522.2Andon Labs Drone-Bench: gpt-5 Localize average run score · Andon Labs
o1 (2024-12-17)13.3Andon Labs Drone-Bench: o1 Localize average run score · Andon Labs
Gemini 2.5 Pro5.8Andon Labs Drone-Bench: gemini-2.5 Localize average run score · Andon Labs
GPT-4o (source unspecified)1.1Andon Labs Drone-Bench: gpt-4o-may Localize average run score · Andon Labs
GPT-4o (2024-11-20)0.0Andon Labs Drone-Bench: gpt-4o-nov Localize average run score · Andon Labs
Loading Atlas data…