Atlas

Benchmarks

CrossSearch (pass^3)

AfterQuery's CrossSearch gives an agent 100 expert-written research requests that can only be answered by combining a task's local files (435 in all) with one to five live web sources the request doesn't name. Agents work in AfterQuery's Terminus 2 harness on Harbor, with one shell that has live internet but no search tool, and attempt each task three times. A GPT-6.1 Sol judge, blind to model identity, checks each answer against 2 to 20 binary rubric items (663 in all), and an attempt earns full credit only when every item passes. The score is pass^3, the share of tasks whose three attempts all earn full credit.

Models
13
Top score
64.0
Leader
Gemini 4 Argon
Updated
Nov 12, 2026
Resolution
~1 mo
Sharp
Saturation
~Jul 2027
Projected
Signal
91stpercentile
195 benchmarks from 0.8 to 3.5224 benchmarks from 3.5 to 6.1133 benchmarks from 6.1 to 8.898 benchmarks from 8.8 to 11.566 benchmarks from 11.5 to 14.126 benchmarks from 14.1 to 16.815 benchmarks from 16.8 to 19.59 benchmarks from 19.5 to 22.18 benchmarks from 22.1 to 24.84 benchmarks from 24.8 to 27.55 benchmarks from 27.5 to 30.12 benchmarks from 30.1 to 32.84 benchmarks from 32.8 to 35.45 benchmarks from 35.4 to 38.11 benchmark from 38.1 to 40.82 benchmarks from 43.4 to 46.11 benchmark from 62.1 to 64.8

73rd strongest of 798

Signal-to-noise +14.98

Difficulty
79thpercentile
1 benchmark from 5.5 to 14.41 benchmark from 14.4 to 23.22 benchmarks from 32.1 to 41.01 benchmark from 41.0 to 49.82 benchmarks from 49.8 to 58.79 benchmarks from 58.7 to 67.67 benchmarks from 67.6 to 76.418 benchmarks from 76.4 to 85.324 benchmarks from 85.3 to 94.241 benchmarks from 94.2 to 103.031 benchmarks from 103.0 to 111.939 benchmarks from 111.9 to 120.850 benchmarks from 120.8 to 129.676 benchmarks from 129.6 to 138.560 benchmarks from 138.5 to 147.4118 benchmarks from 147.4 to 156.2124 benchmarks from 156.2 to 165.1120 benchmarks from 165.1 to 174.039 benchmarks from 174.0 to 182.820 benchmarks from 182.8 to 191.710 benchmarks from 191.7 to 200.61 benchmark from 200.6 to 209.44 benchmarks from 209.4 to 218.3

163rd hardest of 798

Capability needed 166.9

Organizations
13 dated models · 4 frontier modelsClosed weightsOpen weights
Data: AfterQueryatlas.village.dev

CrossSearch (pass^3) leaderboard

#ModelSettingScoreSource
1Gemini 4 Argonhigh · AfterQuery CrossSearch harness: Terminus 2 on Harbor, one shell with live internet, no search tool and a view_image command; three attempts per task; 65,536-token response limit
64.0
2GPT-6 Astrax-high · AfterQuery CrossSearch harness: Terminus 2 on Harbor, one shell with live internet, no search tool and a view_image command; three attempts per task; 128,000-token response limit
60.0
3GPT-6.1 Solx-high · AfterQuery CrossSearch harness: Terminus 2 on Harbor, one shell with live internet, no search tool and a view_image command; three attempts per task; 128,000-token response limit
59.0
4Claude Opus 5.5x-high · AfterQuery CrossSearch harness: Terminus 2 on Harbor, one shell with live internet, no search tool and a view_image command; three attempts per task; 128,000-token response limit
57.0
5Claude Fable 5.2high · AfterQuery CrossSearch harness: Terminus 2 on Harbor, one shell with live internet, no search tool and a view_image command; three attempts per task; 32,768-token response limit
56.0
6Muse Wonderhigh · AfterQuery CrossSearch harness: Terminus 2 on Harbor, one shell with live internet, no search tool and a view_image command; three attempts per task; 32,768-token response limit
42.0
7GPT-5.6 Solx-high · AfterQuery CrossSearch harness: Terminus 2 on Harbor, one shell with live internet, no search tool and a view_image command; three attempts per task; 128,000-token response limit
38.0
8Grok 4.6x-high · AfterQuery CrossSearch harness: Terminus 2 on Harbor, one shell with live internet, no search tool and a view_image command; three attempts per task; 450,000-token response limit
35.0
9Claude Opus 5x-high · AfterQuery CrossSearch harness: Terminus 2 on Harbor, one shell with live internet, no search tool and a view_image command; three attempts per task; 128,000-token response limit
34.0
10Gemini 3.8 Flashhigh · AfterQuery CrossSearch harness: Terminus 2 on Harbor, one shell with live internet, no search tool and a view_image command; three attempts per task; 65,536-token response limit
29.0
11Inkling 2 Smallthinking · AfterQuery CrossSearch harness: Terminus 2 on Harbor, one shell with live internet, no search tool and a view_image command; three attempts per task; 262,144-token response limit; Tinker's 256K-context listing, asked for effort "high", whose numeric setting is undocumented
23.0
12Kimi K3high · AfterQuery CrossSearch harness: Terminus 2 on Harbor, one shell with live internet, no search tool and a view_image command; three attempts per task; 943,718-token response limit
15.0
13Sonnet 5x-high · AfterQuery CrossSearch harness: Terminus 2 on Harbor, one shell with live internet, no search tool and a view_image command; three attempts per task; 128,000-token response limit
7.0

Related benchmarks

Loading Atlas data…