B1K a suite of the exam · Mobile robot
BEHAVIOR-1K
Full household activities in house-scale scenes, for a mobile two-armed robot.
100 full-length household activities in house-scale scenes. Agents must combine high-level reasoning, long-horizon navigation, and dexterous bimanual manipulation. Scored by average task success with BDDL partial credit.
| Origin | adapted from the upstream benchmark (2026 BEHAVIOR Challenge) |
|---|---|
| Simulator | OmniGibson (NVIDIA Isaac Sim) |
| Robot | Galaxea R1 Pro |
| Tasks | 100 tasks · 200 tests |
| Agent time | 120 min |
On a real robot
71 tasks
1
GPT-6.2 AstraCodex · high
54%38/71
2
Claude Fable 5.2Claude Code · high
27%19/71
3
GPT-6 AstraCodex · high
15%11/71
4
GPT-6.1 SolCodex · high
14%10/71
5
Claude Opus 5.5Claude Code · high
6%4/71
6
Gemini 4 ArgonAntigravity · high
4%3/71
7
DeepSeek V4.1 FlashDeepSeek Harness · high
0%0/71
7
Gemini 3.8 FlashAntigravity · high
0%0/71
7
GLM 5.3 FlashClaude Code · high
0%0/71
7
Kimi 3.1 ProClaude Code · high
0%0/71
02
Tasks
100 tasks · 200 tests