SpatialWorld benchmark evaluates interactive spatial reasoning of multimodal agents in real-world tasks
Researchers introduce SpatialWorld, a benchmark for evaluating interactive spatial understanding of multimodal agents across 760 human-annotated tasks spanning household, travel, and social domains. The benchmark integrates eight simulation backends under a shared protocol, requiring agents to operate under vision-only partial observability with egocentric inputs. Evaluation of 15 agents reveals that even the strongest model, GPT-5, achieves only 17.4% task success rate, exposing significant gaps in active exploration and long-horizon planning. The work highlights a mismatch between task success and execution efficiency as a key bottleneck for spatial agents.
Related guides (4)
Related events (8)
ESI-Bench: A Benchmark for Embodied Spatial Intelligence Closing the Perception-Action Loop
ESI-Bench is a new benchmark for embodied spatial intelligence spanning 10 task categories and 29 subcategories, built on OmniGibson and grounded in Spelke's core knowledge systems. It evaluates agents that must actively deploy perception, locomotion, and manipulation to accumulate task-relevant evidence, rather than passively processing oracle observations. Experiments on state-of-the-art MLLMs reveal that active exploration outperforms passive baselines, but most failures stem from 'action blindness'—poor action choices leading to cascading errors—and a metacognitive gap where models commit prematurely with high confidence regardless of evidence quality. Human studies show humans seek falsifying viewpoints and revise beliefs under contradiction, a capability current models lack.
RoboWits: Benchmark for Robotic Creative Problem Solving Under Unexpected Conditions
RoboWits is a new bi-manual robotic benchmark designed to evaluate cognitive reasoning, creative tool use, and robustness to unexpected conditions in robotics. The authors introduce an automated multi-agent task generation pipeline that produces 30 seed tasks and 208 mutated tasks spanning geometry, material, and assembly-based reasoning. Benchmarking results show that pre-trained Vision-Language-Action models (VLAs) achieve limited success on seed tasks after fine-tuning but fail on mutated variants, exposing brittleness in reasoning and strategy adaptation. The benchmark highlights a significant gap between skill-level execution and genuine cognitive reasoning in current robotic systems.
SpatialClaw: Code-as-action interface for agentic 3D/4D spatial reasoning with VLMs
SpatialClaw is a training-free framework that uses code execution as the action interface for vision-language model agents performing spatial reasoning tasks. The system maintains a stateful Python kernel with perception and geometry primitives, allowing the VLM to write iterative executable cells conditioned on prior outputs rather than committing to a full strategy upfront. Evaluated across 20 spatial reasoning benchmarks covering static and dynamic 3D/4D tasks, SpatialClaw achieves 59.9% average accuracy, outperforming the prior state-of-the-art spatial agent by +11.2 points across six VLM backbones.
Benchmark Agent: Autonomous system for end-to-end benchmark construction
Researchers introduce Benchmark Agent, a fully autonomous agentic system that orchestrates the complete benchmark construction pipeline — from query analysis and subtask design to data annotation and quality control. The system was used to produce 15 benchmarks spanning text understanding, multimodal understanding, and domain-specific reasoning, with evaluation via human judges, LLM-as-a-judge, and consistency checks. The work addresses two persistent problems in the field: the labor intensity of benchmark creation and rapid performance saturation after release. Code and a demo will be publicly released.
Imaginative Perception Tokens improve spatial reasoning in vision-language models
Researchers introduce Imaginative Perception Tokens (IPT), intermediate perceptual representations that externalize what a VLM would perceive from alternative spatial viewpoints, enabling reasoning about unobserved spatial structure. The approach is evaluated on three new tasks—Perspective Taking, Path Tracing, and Multiview Counting—using ~20K examples built on the BAGEL backbone. IPT supervision consistently outperforms textual chain-of-thought training for spatial tasks, with the authors finding that forcing spatial computation through language can degrade performance, suggesting a modality mismatch. The work provides both a practical supervision technique and a diagnostic finding about the limits of language-mediated spatial reasoning.
T1-Bench: Multi-scenario agent benchmark across 25 real-world domains
T1-Bench is a new benchmark for evaluating agentic LLM systems in realistic customer-facing, multi-domain environments, covering 25 domains of varying difficulty with interleaved multi-turn scenarios. The authors evaluate 12 proprietary and open-weight models and combine automatic evaluation with human judgments. The benchmark targets gaps in existing agent evals around task complexity, domain diversity, and compositional reasoning across multi-step interactions.
M³Exam: Benchmark for Multimodal Memory in Realistic User-Agent Interactions
Researchers introduce M³Exam, a query-centric multimodal conversational memory benchmark designed to evaluate language agents on realistic user-agent interactions, including cross-modal grounding and implicit information inference. Existing benchmarks are critiqued for assuming sparse visuals and human-human interaction formats. The paper also proposes M³Proctor, a companion memory method that detects query modality bias and retrieves raw visual sources on demand, achieving 13% accuracy improvement while reducing index-construction time and retrieved tokens by over 70%.
STT-Arena: Benchmark for Adaptive Replanning Under Spatio-Temporal Dynamics in Tool-Using LLMs
STT-Arena is a new benchmark of 227 interactive tasks designed to evaluate LLMs' ability to detect mid-task disruptions and replan under spatio-temporal dynamics, covering nine conflict types and four solvability levels. Evaluation of frontier models including Claude-4.6-Opus shows less than 40% overall accuracy, revealing fundamental limitations in dynamic reasoning. The authors identify three recurring failure modes—Stale-State Execution, Misdiagnosis of Dynamic Triggers, and Missing Post-Adaptation Verification—and propose an iterative trajectory refinement technique combined with online RL to train STT-Agent-4B, a 4B-parameter model that outperforms frontier LLMs on the benchmark.