GPT-6-Astra demonstrates a level of competence in robotic manipulation and navigation that challenges the conventional paradigm of building specialized models for individual tasks and embodiments. In this report, we ask not only how capable GPT-6-Astra is, but what its capabilities imply for the field. First, we probe the limits of standalone GPT-6-Astra. We conduct a large-scale evaluation across 12 representative benchmarks spanning tabletop manipulation, humanoids, mobile platforms, and dexterous hands. Under a unified protocol that allows in-context learning (ICL) from training-set demonstrations and substantially increases the number of execution steps, GPT-6-Astra achieves unprecedented performance (Fig. 1). Future research may focus on improving 1) the generality of ICL 2) cost and execution time, 3) physical-state verification, 4) recovery from irreversible failures, and 5) long-horizon preservation of progress. Second, we study the agentic harness for embodied control. We identify that close-loop coordination between a high-level "System 2" reasoner (i.e., GPT-6-Astra) and a low-level "System 1" controller (e.g., π0.5) is costly but effective. Third, we examine GPT-6-Astra on real-world deployment. Our experiments on 8 robot configurations reveal substantial gaps: 1) robot demonstrations required for ICL are expensive to collect, and 2) it is implausible to trade longer execution for higher success rates with physical constraints. Fourth, we envision the next generation of embodied benchmarks. We develop a prototype that 1) derives tasks from real-world needs, 2) provides no in-domain demonstrations, and 3) judges agents on a world clock that keeps running while they deliberate. Taken together, our results suggest that GPT-6-Astra has largely crossed the threshold of embodied competence, but not yet that of practical embodied intelligence.
Figures & tables
Figure 1: Probing the capability upper bound of GPT-6-Astra across 12 benchmarks under a unified protocol that 1) allows for in-context learning (ICL) on the training set and 2) extends the step limit (budget) up to 10× . RoboLab has no training set and thus no ICL; BEHAVIOR is not extended to 10× due to its large official budget. SOTA cut-off date: October 12, 2026. Details in Tab. 2 .
Figure 2: In-context learning protocol. For each task, GPT-6-Astra studies the training data and writes a knowledge package K , which is frozen before testing. Each test episode starts a fresh session with K and runs under the standard ( 1× ) or extended ( 10× ) budget. Model weights remain fixed throughout, and K remains unchanged during testing, and no information passes between episodes.
Figure 3: Contents of an ICL knowledge package. GPT-6-Astra decides what to retain, yet packages share six recurring components; file types are typical and content varies by benchmark. At test time, K is read alongside current observations to retrieve, adapt and check actions.
Figure 4: Success rate versus step budget. The budget is expressed as a multiple of each benchmark’s official step limit; dots show observed success rates and pink curves show fitted trends. We evaluate budgets up to 10× the official limit ( 1× for BEHAVIOR). The gray line marks the upper bound, which assumes that every episode that exhausts the evaluated budget without succeeding has the potential to succeed with an unlimited budget.
Figure 5: (a) Average per-task token usage and (b) average per-task time for each benchmark, stacked from bottom to top as ICL preparation, the 1× run, and the additional cost of extending to 10× . Here, 1× and 10× denote maximum action-step budgets, not token or time budgets. The relatively small additional token and time costs primarily reflect early termination of extended runs upon success or terminal failure, before the full 10× budget is exhausted. We observe that the vast majority of tokens consumed are cached input tokens (approximately 98% in our runs). Accordingly, we use a simple rule of thumb for cost estimation: 1M tokens ≈ 1 USD .
Figure 6: Residual failures under ICL with the extended budget. Each pair shows the state before a key event and its outcome; purple arrows mark the relevant objects. (a) Verifying physical success: apparent or self-reported completion diverges from the actual task state. (b) Recovering from consequential actions: an early mishap, such as a spill or a dropped part, cannot be undone and derails later subtasks. (c) Preserving progress over long horizons: progress that later steps depend on, such as an open lid or a placed object, is lost and blocks the task.
Figure 7: Token cost versus performance on RoboDojo for different System 1–System 2 collaboration strategies and action interfaces. Each point represents an experimental configuration, with token cost measured in millions and performance measured by the unweighted mean of category success rate. Dashed lines intersect at the Agentic GPT-6-Astra baseline reference point, defining four shaded quadrants.
Embodiment
#
Main question
Tasks
Single arm, gripper Franka 3, Franka Panda, PiPER + Pika gripper
3
Do simulation results transfer, from rigid to deformable and precise tasks?
Stacking and container nesting; cloth and towel folding; plug insertion
Dual arm, grippers AgileX Piper
1
Can GPT-6-Astra coordinate two arms on harder zero-shot tasks?
Wheeled mobile manipulator Galaxea R1 Pro, AgileX Tracer + xArm7
2
Can GPT-6-Astra act over room-scale distances and long horizons?
Find washer and press button; flower collection; tray arrangement; navigation and search
Humanoid Unitree G1
1
Can GPT-6-Astra act as System 2 over low-level locomotion and arm control?
Tabletop pick-and-drop; turn, approach and stop
Table 1: Real-world platforms, ordered from simple to complex embodiments. Each class probes a different question; settings and per-task results are in App. 11 .
Figure 8: Zero-shot real-world control, from simple to complex embodiments. Frames from released videos in temporal order. Chips give successes over all counted trials. (c) GPT-6-Astra re-plans after contact displaces the receiving bowl. (e) Dexterous grasping with the Wuji hands of the Tianji M6S, given human-written grasp scripts (the successful bottle run; the toy grasp succeeds in 3/3). The full gallery is in Fig. 13 , and all tasks per platform in App. 11 .
Figure 9: What GPT-6-Astra can do depends on what it can see. Flower insertion, main view and wrist views. (a) Single arm (Franka 3): the gripper holds the stem vertically, so its wrist camera cannot see the vase mouth, and the insertion fails. (b) Dual-arm Piper: the wrist view of the holding arm shows only the stem; GPT-6-Astra moves the free arm until its wrist camera sees the vase mouth (boxed frames: no vase, approaching the vase, vase mouth in view), and the insertion succeeds.
Figure 11: Building and using the household environment. Publicly available assets (1) and a floor plan (2) are composed into an interactive home. Coding agents place everyday items at real-world scale on named supports under human review (3), and every part is bound to physics: articulation, liquids, cloth and contact (4). A task states only a need (5): asked to make breakfast, the agent must work out the fridge, air fryer and toaster by itself, and is to be judged both on the request (Intent) and on obligations no one stated, such as closing the fridge (Duty). The agent sees only through the robot’s onboard cameras, and its deliberation advances world time (6).
Figure 12: GPT-6-Astra in the household proof of concept. All runs are zero-shot, with contact-based grasping and no demonstrations. (a) While moving a book during tidying, the robot displaces desk items, knocks objects onto the floor and pushes the monitor out of place with its elbow. (b) Having opened the refrigerator to store the milk, the robot turns to carrying fruit to the table, and the door stays open from 300 s to 765 s of world time. (c) Making breakfast, the robot holds the refrigerator door open with one arm, but from where it stands the door blocks its other arm from the bacon inside; when it releases the handle to reach in, the door swings shut. (d) The running tap is not closed, and with the drain closed the sink fills until water spills down the cabinet onto the floor. (e) GPT-6-Astra watches an egg through the robot’s head camera and decides to turn the burner off once it has set; deciding takes 8 s of world time with the burner still on, and by the time the burner is turned off, about 90 s after the egg has set, the egg has charred. Purple outlines highlight relevant objects; timestamps denote elapsed world time. External camera views are used for visualization; the agent sees only through the robot’s onboard cameras.
Embodiment / benchmark
Published baselines
GPT-6-Astra (ours)
SOTA
Agentic SOTA
Up to 1×
Up to 10×
Method
SR
Method
SR
Cutoff at 10×
Cutoff at 1×
Tabletop robotic arms
LIBERO-Long++ [ 162 ]
LiLo-VLA [ 162 ]
82.0
–
–
97.5
100.0
97.5
LIBERO-PRO [ 183 ]
πRLinf [ 175 ]
50.0
RPent / GPT-6 Astra [ 175 ]
92.6
98.3
99.3
98.3
LIBERO-Ultra-Long [ 162 ]
LiLo-VLA [ 162 ]
44.0
–
–
82.2
98.9
82.2
Table 2: Benchmark-level success rates (%) underlying Fig. 1 . For each benchmark, we list the best published result (SOTA) with its method, the best published agentic method (Agentic SOTA), and GPT-6-Astra + ICL under the standard ( 1× ) and cumulative extended (up to 10× ) budgets. Benchmarks are grouped by embodiment. Published baselines are reported under the standard budget. The 1× execution in this table is performed separately from 10× , with the respective budget included in the context during execution. Dashes mean no published agentic result or no extended-budget run. ∗ denotes the result with VLA model, while ‡ denotes the zero-shot results without access to the training data.
Policy
Type
SR (%)
Size
GPT-6-Astra 10x
Agent
69.2
–
GPT-6-Astra 1x
Agent
57.1
–
FLUX 3 Action [ 12 ]
WAM
42.9
7B
Cosmos3-Nano-Policy [ 116 ]
WAM
36.8
16B
π0.5 [ 13 ]
VLA
28.0
3.3B
DreamZero [ 166 ]
WAM
25.7
14B
Table 6: Overall performance on RoboLab. Success rate (SR) is reported in percent; higher is better. Policies are ordered by success rate, and the best result is shown in bold.
Policy
Type
SR
ISR
CSR
CGT
Size
Tokens
GPT-6-Astra Max + ICL 10×
Agent
28.95
62.87
52.97
52.58
–
2.587
GPT-6-Astra Max + ICL 1×
Agent
21.05
54.68
44.55
43.91
–
1.317
Claude Code Opus-5
Agent
5.26
26.87
17.82
17.84
–
–
FantasyVLN [ 186 ]
VLN
2.44
11.01
9.64
8.99
7B
–
Aux-think [ 147 ]
VLN
0.65
3.16
2.04
1.47
8B
–
MGDM [ 134 ]
VLN
0.00
2.35
1.65
2.92
7B
–
Table 9: Overall performance on LHPR-VLN [ 134 ] . SR, ISR, CSR, and CGT are percentages, with higher values indicating better performance. Token consumption is reported in millions. Policies are ordered by SR. Bold values mark the highest navigation scores.
Policy
Size
Type
Observation
R2R val-unseen
RxR val-unseen
NE ↓
OS ↑
SR ↑
SPL ↑
NE ↓
SR ↑
SPL ↑
nDTW ↑
GPT-6-Astra Max + ICL 10×
–
Agent
Mono. RGB
2.34
91
88
76.98
1.36
90
73.52
82.98
GPT-6-Astra Max + ICL 1×
–
Agent
Mono. RGB
3.45
84
80
68.47
2.08
82
65.61
77.85
GPT-6-Astra Ultra [ 33 ]
–
Agent
Mono. RGB
2.9
83.7
81.3
71.5
–
–
–
–
Robostral Navigate [ 16 ]
8B + 121M
VLN
Mono. RGB
3.2
81.3
77.4
74.2
3.47
75.1
68.7
–
Qwen-RobotNav-8B [ 172 ]
8B
VLN
Pano. RGB
3.53
78.5
72.1
66.6
3.58
76.5
65.7
72.5
Table 10: Overall performance on VLN-CE [ 87 ] R2R and RxR val-unseen. OS, SR, SPL, and nDTW are percentages, and NE is in meters. Bold values mark the best reported navigation scores.
Metric
GPT-6-Astra Zero-shot
GPT-6-Astra ICL
Psi-0 [ 151 ] (ref.)
Success rate
Common-4
85/120 (70.8%)
109/120 (90.8%)
108/120 (90.0%)
Six-task (reference only) c
110/167 (65.9%)
111/126 (88.1%)
154/180 (85.6%)
Per-episode efficiency on matched successes a
Mean action steps
376.9
210.4
–
Mean test tokens (millions)
2.64
6.40
–
Table 12: GPT-6-Astra standard-budget ( 1× ) success, efficiency and cost on SIMPLE: zero-shot vs. ICL. Common-4 covers all 120 episodes from XMovePick, BendPick, Handover and Tabletop; it differs from the 80-episode Matched-8 subset in Table 11 . Efficiency is measured on the 81 episodes solved in both settings. Six-task success is provided for reference only. Psi-0 is the published baseline.
GPT-6 Astra exhibits a remarkable ability to generate numerical robot actions, extending its role beyond high-level planning. To assess Astra's capabilities as general-purpose embodied policies, we conduct comprehensive evaluations across six domains, examining direct control, cooperation with learned policies, and feedback-driven adaptation. In gripper manipulation, Astra can correct task targets and prepare contact conditions for subsequent policy execution; hybrid control with π0.5 achieves 48% success on the evaluated RoboDojo subset. In dexterous manipulation, hybrid control achieves 50% success in ten experience-guided DexJoCo trials, while direct in-hand control struggles to coordinate finger contacts. In mobile manipulation, hybrid control reaches 38.7% success on the evaluated RoboCasa365. In navigation, Astra leads our local comparisons, reaching 92% success on RxR instruction following and 82% on HM3D object search, although search incurs substantial detours. In locomotion, dense motion-reference generation remains unreliable: none of five sequential attempts on a single obstacle course reaches the goal, despite improvements in stability and forward progress. In humanoid loco-manipulation, Astra exceeds baseline methods on 13 of 30 HumanoidBench tasks with pretrained whole-body controllers. These findings reveal a gap between useful task decisions and reliable physical control. Inference latency further constrains practical control: across 50 RoboDojo instances per condition, policy-assisted and direct control consume 624.8 million and 1.132 billion tokens. A 30-second locomotion run requires 250 model calls averaging 39.86 seconds each, with physics paused during inference.
Embodied AI systems are often organized into System 1 and System 2. System 1 is typically a pretrained policy that generates actions at high frequency, whereas System 2 is often instantiated as a vision-enabled language model for high-level planning. We ask whether a large language model (LLM) can act as the policy for robot manipulation without task-specific finetuning. We call this setting LLM as policy. We evaluate three LLMs on all 42 RoboDojo tasks and compare their scores with 40 public policies. Astra and GPT-5.5 use the official 50-episode-per-task protocol; DeepSeek-Flash uses 10 episodes per task. GPT-6 Astra achieves 22.48% average success rate and 28.97 Score over 2,100 trials, ranking above every public entry. Yet GPT-5.5 and DeepSeek-Flash reach only 0.88% and 1.92% average success rate with the same post-processing. We find that Astra exhibits a sharply polarized capability profile. It generalizes well to tasks that require semantic understanding but not high-precision control. In contrast, it performs poorly on tasks that require precision, dynamic control, or complex bimanual coordination. In-context experiments show no aggregate benefit from one-shot demonstrations, while selected interaction traces show within-episode corrections under perturbations. Overall, the evaluated LLMs vary substantially in manipulation performance. Astra stands out and provides initial evidence for the potential of a general-purpose manipulation model, although reliable precision and dynamic control remain limitations in the evaluated setting.
Wenbo Zhang, Kaixuan Wang, Yutao Ouyang +9
RoboProbe · RoboDojo · The University of Hong Kong +5
General-purpose multimodal agents can write robot-control programs, but repeated exploration and model-mediated action selection can make execution slow. We study how external body knowledge, successful experience, and executable skills improve an XLeRobot controlled by GPT-6-Astra in a simulated and a physical elevator-button task. In 30 fixed-start simulation trials, complete robot geometry and camera information reduce mean completion time by 57.4% relative to a baseline with only the common control interface and no prior experience; images with synchronized action and state records reduce it by 68.6% without additional body assets. In nine paired comparisons (18 trials) at starts displaced by 10-100 cm, experience recorded at the original start reduces mean time by 58-63% relative to no experience, demonstrating generalization to the tested new starting positions. During experience experiments, GPT-6-Astra spontaneously generates a short visual-feedback program. Researcher-refactored versions reduce mean local-task time by 29-31% in 27 simulation trials. Finally, 12 real-robot trials using operator-confirmed button contact demonstrate sim2real reuse: at a shared nominal start, simulation XML assets and simulation experience reduce mean time by 53.0% and 49.9%, respectively; real experience also transfers to two new starts. These results suggest a practical way to build general-purpose manipulation experiments around GPT-6-Astra: supply machine-readable body descriptions and synchronized demonstrations, and turn useful agent-generated feedback routines into reusable skills, while the agent adapts actions from current images. We release all task prompts, trial-level experimental data, and acquired skill implementations at https://github.com/hesd10/astra-robot-sim2real.