Embodied Agent Interface
Models interpret goals, decompose them into subgoals, sequence actions, and predict state transitions in BEHAVIOR and VirtualHome. The evaluator checks task completion and distinguishes hallucinations, invalid actions, and planning errors.
The four modules have separate metrics. Running the BEHAVIOR track requires iGibson assets.
- Made by
- Stanford Vision and Learning Lab
- Released
- October 2024
- Kind
- Embodied reasoning
- Level
- Everyday
- Scored
- Checked automatically
- Answers
- Agent task
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Agentic
- Environment
- VirtualHome, BEHAVIOR
- Items from
- Another benchmark, Generated
- Refresh
- One fixed set
- Run by
- Anyone
- Based on
- BEHAVIOR-1K
- Strongest model tested
- o1-mini #142 on the capability index
- Last model added
- September 2024
- Scores refreshed
- November 20th, 2026
In the authors’ words
We aim to evaluate Large Language Models (LLMs) for embodied decision making. While a significant body of work has been leveraging LLMs for decision making in embodied environments, we still lack a systematic understanding of their performance because they are usually applied in different domains, for different purposes, and built based on different inputs and outputs. Furthermore, existing evaluations tend to rely solely on a final success rate, making it difficult to pinpoint what ability is missing in LLMs and where the problem lies, which in turn blocks embodied agents from leveraging LLMs effectively and selectively. To address these limitations, we propose a generalized interface (Embodied Agent Interface) that supports the formalization of various types of tasks and input-output specifications of LLM-based modules. Specifically, it allows us to unify 1) a broad set of embodied decision-making tasks involving both state and temporally extended goals, 2) four commonly-used LLM-based modules for decision making: goal interpretation, subgoal decomposition, action sequencing, and transition modeling, and 3) a collection of fine-grained metrics which break down evaluation into various types of errors, such as hallucination errors, affordance errors, various types of planning errors, etc. Overall, our benchmark offers a comprehensive assessment of LLMs' performance for different subtasks, pinpointing the strengths and weaknesses in LLM-powered embodied AI systems, and providing insights for effective and selective use of LLMs in embodied decision making.
o1-preview leads with 70.3 among models released by September 2024 o1-preview leads with 70.3 among models released by September 2024
Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. The leaderboard gives a separate overall score for each environment, so the headline averages BEHAVIOR and VirtualHome equally. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.
The newest model with a score here came out in September 2024, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 70.3 o1-preview
- 64.5 Claude 3.5 Sonnet
- 60.3 GPT-4o
- 60.1 Claude 3 Opus
- 57.7 o1-mini
- 57.0 Gemini 1.5 Pro (snapshot unspecified)
- 54.0
Gemini 1.5 Flash
- 53.0 GPT-4 Turbo
- 52.9 Claude 3 Sonnet
- 52.5
Mistral Large
- 47.3 Llama3 70B Instruct
- 45.0 Mixtral 8x22B MoE
- 44.3 Cohere Command R+
- 44.0 Claude 3 Haiku
- 37.2
Gemini 1.0 Pro
- 33.0
Command R (03-2024)
- 31.8 GPT-3.5 Turbo
- 25.8 Llama3 8B Instruct
- 1 o1-preview 70.3
- 2 Claude 3.5 Sonnet 64.5
- 3 GPT-4o May 2024 60.3
- 4 Claude 3 Opus 60.1
- 5 o1-mini 57.7
- 6 Gemini 1.5 Pro (snapshot unspecified) 57.0
- 7
Gemini 1.5 Flash May 2024 54.0
- 8 GPT-4 Turbo Apr 2024 53.0
- 9 Claude 3 Sonnet 52.9
- 10
Mistral Large 52.5
- 11 Llama3 70B Instruct 47.3
- 12 Mixtral 8x22B MoE 45.0
- 13 Cohere Command R+ 44.3
- 14 Claude 3 Haiku 44.0
- 15
Gemini 1.0 Pro 37.2
- 16
Command R (03-2024) 33.0
- 17 GPT-3.5 Turbo Jan 2024 31.8
- 18 Llama3 8B Instruct 25.8
Source: Official Embodied Agent Interface leaderboard
- 1 o1-preview 70.3
- 2 Claude 3.5 Sonnet 64.5
- 3 GPT-4o May 2024 60.3
- 4 Claude 3 Opus 60.1
- 5 o1-mini 57.7
- 6
Gemini 1.5 Flash May 2024 54.0
- 7 GPT-4 Turbo Apr 2024 53.0
- 8 Claude 3 Sonnet 52.9
- 9
Mistral Large 52.5
- 10 Claude 3 Haiku 44.0
- 11
Gemini 1.0 Pro 37.2
- 12
Command R (03-2024) 33.0
- 13 GPT-3.5 Turbo Jan 2024 31.8
Source: Official Embodied Agent Interface leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | o1-preview | 70.3 | Sep 2024 | Official leaderboard | |
| 2 | Claude 3.5 Sonnet | 64.5 | Jun 2024 | Official leaderboard | |
| 3 | GPT-4o May 2024 | 60.3 | May 2024 | Official leaderboard | |
| 4 | Claude 3 Opus | 60.1 | Feb 2024 | Official leaderboard | |
| 5 | o1-mini | 57.7 | Sep 2024 | Official leaderboard | |
| 6 | Gemini 1.5 Pro (snapshot unspecified) closed weights | 57.0 | Official leaderboard | ||
| 7 | | 54.0 | May 2024 | Official leaderboard | |
| 8 | GPT-4 Turbo Apr 2024 | 53.0 | Apr 2024 | Official leaderboard | |
| 9 | Claude 3 Sonnet | 52.9 | Feb 2024 | Official leaderboard | |
| 10 | | 52.5 | Feb 2024 | Official leaderboard | |
| 11 | Llama3 70B Instruct open weights | 47.3 | Official leaderboard | ||
| 12 | Mixtral 8x22B MoE open weights | 45.0 | Official leaderboard | ||
| 13 | Cohere Command R+ open weights | 44.3 | Official leaderboard | ||
| 14 | Claude 3 Haiku | 44.0 | Mar 2024 | Official leaderboard | |
| 15 | | 37.2 | Dec 2023 | Official leaderboard | |
| 16 | | 33.0 | Mar 2024 | Official leaderboard | |
| 17 | GPT-3.5 Turbo Jan 2024 | 31.8 | Jan 2024 | Official leaderboard | |
| 18 | Llama3 8B Instruct open weights | 25.8 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | o1-preview | 70.3 | Sep 2024 | Official leaderboard | |
| 2 | Claude 3.5 Sonnet | 64.5 | Jun 2024 | Official leaderboard | |
| 3 | GPT-4o May 2024 | 60.3 | May 2024 | Official leaderboard | |
| 4 | Claude 3 Opus | 60.1 | Feb 2024 | Official leaderboard | |
| 5 | o1-mini | 57.7 | Sep 2024 | Official leaderboard | |
| 6 | | 54.0 | May 2024 | Official leaderboard | |
| 7 | GPT-4 Turbo Apr 2024 | 53.0 | Apr 2024 | Official leaderboard | |
| 8 | Claude 3 Sonnet | 52.9 | Feb 2024 | Official leaderboard | |
| 9 | | 52.5 | Feb 2024 | Official leaderboard | |
| 10 | Claude 3 Haiku | 44.0 | Mar 2024 | Official leaderboard | |
| 11 | | 37.2 | Dec 2023 | Official leaderboard | |
| 12 | | 33.0 | Mar 2024 | Official leaderboard | |
| 13 | GPT-3.5 Turbo Jan 2024 | 31.8 | Jan 2024 | Official leaderboard |
Rankings by part
We read 16 parts from the benchmark’s own leaderboard in 2 groups: behavior and virtualhome. Each ranks models on one part of the test. o1-preview leads 10 of them, and 3 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
Overall Performance
13 models we track have a score here.
- 1 o1-preview 74.9
- 2 Claude 3.5 Sonnet 64.2
- 3 Claude 3 Opus ↑1 60.4
- 4 GPT-4o May 2024 ↓1 59.8
- 5 o1-mini 57.5
- 6 Claude 3 Sonnet ↑2 55.1
- 7
Gemini 1.5 Flash May 2024 ↓1 52.1
- 8
Mistral Large ↑1 50.4
- 9 GPT-4 Turbo Apr 2024 ↓2 49.6
- 10 Claude 3 Haiku 41.6
Goal Interpretation F1
13 models we track have a score here.
- 1 Claude 3.5 Sonnet ↑1 82.7
- 2 o1-preview ↓1 81.6
- 3 GPT-4o May 2024 79.2
- 4 GPT-4 Turbo Apr 2024 ↑3 77.2
- 5 Claude 3 Opus ↓1 77.0
- 6 o1-mini ↓1 76.4
- 7
Gemini 1.5 Flash May 2024 ↓1 74.8
- 8
Mistral Large ↑1 74.3
- 9 Claude 3 Sonnet ↓1 69.4
- 10
Gemini 1.0 Pro ↑1 60.0
Action Sequencing Task SR
13 models we track have a score here.
- 1 o1-preview 81.0
- 2 Claude 3.5 Sonnet 60.0
- 3 o1-mini ↑2 56.0
- 4 Claude 3 Opus 51.0
- 5 GPT-4o May 2024 ↓2 47.0
- 6 Claude 3 Sonnet ↑2 44.0
- 7
Gemini 1.5 Flash May 2024 ↓1 40.0
- 8 GPT-4 Turbo Apr 2024 ↓1 38.0
- 9
Mistral Large 33.0
- 10
Gemini 1.0 Pro ↑1 27.0
Action Sequencing Execution SR
13 models we track have a score here.
- 1 o1-preview 91.0
- 2 Claude 3.5 Sonnet 69.0
- 3 o1-mini ↑2 65.0
- 4 Claude 3 Opus 59.0
- 5 Claude 3 Sonnet ↑3 57.0
- 6 GPT-4o May 2024 ↓3 53.0
- 7
Gemini 1.5 Flash May 2024 ↓1 52.0
- 8
Mistral Large ↑1 50.0
- 9 GPT-4 Turbo Apr 2024 ↓2 45.0
- 10
Gemini 1.0 Pro ↑1 32.0
Subgoal Decomposition Task SR
13 models we track have a score here.
- 1 o1-preview 57.0
- 2 GPT-4o May 2024 ↑1 49.0
- 3 Claude 3 Opus ↑1 41.0
- 4 Claude 3 Sonnet ↑4 39.0
- 5 Claude 3.5 Sonnet ↓3 39.0
- 6 GPT-4 Turbo Apr 2024 ↑1 38.0
- 7
Gemini 1.5 Flash May 2024 ↓1 34.0
- 8
Mistral Large ↑1 31.0
- 9 o1-mini ↓4 31.0
- 10 Claude 3 Haiku 30.0
Subgoal Decomposition Execution SR
13 models we track have a score here.
- 1 o1-preview 62.0
- 2 GPT-4o May 2024 ↑1 55.0
- 3 Claude 3 Opus ↑1 47.0
- 4 GPT-4 Turbo Apr 2024 ↑3 47.0
- 5 Claude 3.5 Sonnet ↓3 44.0
- 6 Claude 3 Sonnet ↑2 43.0
- 7
Gemini 1.5 Flash May 2024 ↓1 42.0
- 8 o1-mini ↓3 39.0
- 9
Mistral Large 38.0
- 10 GPT-3.5 Turbo Jan 2024 ↑3 36.0
Transition Modeling F1
13 models we track have a score here.
- 1 o1-preview 70.8
- 2 Claude 3.5 Sonnet 67.9
- 3 Claude 3 Opus ↑1 63.4
- 4 GPT-4o May 2024 ↓1 60.9
- 5 o1-mini 56.4
- 6 Claude 3 Sonnet ↑2 56.2
- 7
Gemini 1.5 Flash May 2024 ↓1 53.4
- 8 Claude 3 Haiku ↑2 51.6
- 9
Mistral Large 49.5
- 10
Gemini 1.0 Pro ↑1 45.8
Transition Modeling Planner SR
13 models we track have a score here.
- 1 o1-preview 89.0
- 2 Claude 3 Opus ↑2 82.0
- 3 Claude 3.5 Sonnet ↓1 82.0
- 4 Claude 3 Sonnet ↑4 80.0
- 5
Mistral Large ↑4 77.0
- 6 o1-mini ↓1 77.0
- 7 GPT-4o May 2024 ↓4 67.0
- 8
Gemini 1.5 Flash May 2024 ↓2 66.0
- 9 Claude 3 Haiku ↑1 64.0
- 10 GPT-4 Turbo Apr 2024 ↓3 46.0
Overall Performance
13 models we track have a score here.
- 1 o1-preview 65.8
- 2 Claude 3.5 Sonnet 64.9
- 3 GPT-4o May 2024 60.8
- 4 Claude 3 Opus 59.9
- 5 o1-mini 57.9
- 6 GPT-4 Turbo Apr 2024 ↑1 56.3
- 7
Gemini 1.5 Flash May 2024 ↓1 56.0
- 8
Mistral Large ↑1 54.5
- 9 Claude 3 Sonnet ↓1 50.7
- 10 Claude 3 Haiku 46.5
Goal Interpretation F1
13 models we track have a score here.
- 1 o1-preview 42.7
- 2
Command R (03-2024) ↑10 36.7
- 3 GPT-4o May 2024 36.5
- 4 GPT-4 Turbo Apr 2024 ↑3 33.2
- 5 Claude 3.5 Sonnet ↓3 33.0
- 6 Claude 3 Opus ↓2 31.4
- 7 o1-mini ↓2 31.2
- 8 Claude 3 Sonnet 29.4
- 9 Claude 3 Haiku ↑1 28.0
- 10
Mistral Large ↓1 26.8
Action Sequencing Task SR
13 models we track have a score here.
- 1
Mistral Large ↑8 73.4
- 2 Claude 3.5 Sonnet 72.8
- 3 o1-preview ↓2 71.1
- 4 Claude 3 Opus 66.2
- 5 o1-mini 65.9
- 6 Claude 3 Sonnet ↑2 62.9
- 7
Gemini 1.5 Flash May 2024 ↓1 61.9
- 8 GPT-4o May 2024 ↓5 61.6
- 9 GPT-4 Turbo Apr 2024 ↓2 57.0
- 10 Claude 3 Haiku 43.3
Action Sequencing Execution SR
13 models we track have a score here.
- 1
Mistral Large ↑8 83.6
- 2 o1-preview ↓1 78.4
- 3 Claude 3.5 Sonnet ↓1 75.4
- 4 GPT-4o May 2024 ↓1 71.1
- 5 Claude 3 Opus ↓1 70.8
- 6 o1-mini ↓1 68.9
- 7 Claude 3 Sonnet ↑1 67.2
- 8
Gemini 1.5 Flash May 2024 ↓2 67.2
- 9 GPT-4 Turbo Apr 2024 ↓2 65.6
- 10 Claude 3 Haiku 48.5
Subgoal Decomposition Task SR
13 models we track have a score here.
- 1 o1-preview 89.4
- 2
Gemini 1.5 Flash May 2024 ↑4 89.1
- 3 Claude 3.5 Sonnet ↓1 89.1
- 4 GPT-4o May 2024 ↓1 87.6
- 5 Claude 3 Opus ↓1 86.7
- 6 GPT-4 Turbo Apr 2024 ↑1 85.5
- 7
Mistral Large ↑2 84.3
- 8 Claude 3 Sonnet 83.1
- 9 o1-mini ↓4 79.3
- 10 Claude 3 Haiku 78.4
Subgoal Decomposition Execution SR
13 models we track have a score here.
- 1 GPT-4 Turbo Apr 2024 ↑6 94.1
- 2
Gemini 1.5 Flash May 2024 ↑4 94.1
- 3 o1-preview ↓2 93.2
- 4
Mistral Large ↑5 92.0
- 5 Claude 3.5 Sonnet ↓3 92.0
- 6 GPT-4o May 2024 ↓3 91.1
- 7 Claude 3 Opus ↓3 89.9
- 8 Claude 3 Sonnet 86.4
- 9
Gemini 1.0 Pro ↑2 84.6
- 10 o1-mini ↓5 84.6
Transition Modeling F1
13 models we track have a score here.
- 1 Claude 3.5 Sonnet ↑1 48.9
- 2 Claude 3 Opus ↑2 48.8
- 3 o1-preview ↓2 48.0
- 4 GPT-4o May 2024 ↓1 46.7
- 5
Gemini 1.5 Flash May 2024 ↑1 45.7
- 6 GPT-4 Turbo Apr 2024 ↑1 42.9
- 7 Claude 3 Haiku ↑3 42.3
- 8
Gemini 1.0 Pro ↑3 41.8
- 9 o1-mini ↓4 41.5
- 10 Claude 3 Sonnet ↓2 41.2
Transition Modeling Planner SR
13 models we track have a score here.
- 1 Claude 3.5 Sonnet ↑1 80.5
- 2 o1-preview ↓1 72.4
- 3 o1-mini ↑2 69.0
- 4 GPT-4o May 2024 ↓1 68.2
- 5 Claude 3 Opus ↓1 61.8
- 6 GPT-4 Turbo Apr 2024 ↑1 56.1
- 7
Command R (03-2024) ↑5 51.1
- 8
Gemini 1.5 Flash May 2024 ↓2 46.6
- 9
Mistral Large 31.1
- 10 Claude 3 Haiku 30.4
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official Embodied Agent Interface leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official Embodied Agent Interface leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in Robotics
- ERQA In use 400 multiple-choice visual questions from robot and egocentric footage on spatial, trajectory, action and state reasoning. Accuracy.
Gemini Robotics ER 2 78.5%
- SimplerEnv In use Real-to-sim replicas of Google Robot and WidowX/Bridge setups. Policies scored by success rate on pick, move, drawer and Bridge tasks. Praxis-1 80%
- RoboLab In use Zero-shot DROID Franka policies in Isaac Lab on 120 tabletop tasks at three instruction levels, 10 episodes each. Success rate. Praxis-1 46.2%
- RoboDojo In use Generalist manipulation policies run 42 Isaac Sim tasks and 18 remote real-robot tasks. Scored by task score and success rate. GPT-6 Astra 27.6%
- Progress classification (Google DeepMind) In use Models assign each frame of a robot video to one of five task-progress bands from 0 to 100 percent. Scored by accuracy.
Gemini Robotics ER 2 57.4%
- Human proximity (Google DeepMind robotics) In use Models judge from robot camera views whether a human is within 1 meter, a spatial awareness check for safe robot behavior. Scored by accuracy.
Gemini Robotics ER 2 93%
Data audit
8 of 9 checks pass, 1 warning. Look first at the Newest score check. The newest came 2 years ago, in September 2024, for o1-mini. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 1 warning 2 pass 2 notes
- Pass Independent runs Scores from the official leaderboard (18 entries).
- Pass Models scored 13 models we track, plus 5 other entries on its leaderboard.
- Warning Newest score The newest came 2 years ago, in September 2024, for o1-mini.
- Note Frontier models None of the top 10 has a score. The strongest tested is o1-mini (#142).
- Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
- Pass Importer We read the board on embodied-agent-interface.github.io: 13 rows for models we track and 5 other entries at the last read.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.