The China Merchants Lion Mountain Lab released the open-source embodied reasoning VLM 'Lion-ER-35B-A3B'. Utilizing a MoE architecture (35B total, 3B active), it shifts from data accumulation to a 'capability-driven' data engine, decomposing tasks into 8 domains and 39 fine-grained capabilities for iterative optimization. The model achieves improvements across ten mainstream embodied benchmarks, surpassing larger models in key areas by enhancing robots' understanding of 3D space, object states, and task progress.
Embodied Reasoning
具身推理
The ability of an agent to perceive, decide, and act based on interaction between its body and environment.
Curated Intelligence
The Lion Mountain AI Lab under China Merchants Advanced Technology Institute has released Lion-ER-35B-A3B, an embodied-reasoning vision-language model built on a mixture-of-experts architecture with roughly 35B total parameters and only about 3B activated per token. The lab claims across-the-board gains on ten embodied-AI benchmarks, with some capabilities surpassing larger models, and attributes this to a capability-first data engine that decomposes embodied reasoning into 8 top-level domains and 39 sub-capabilities and iterates data against evaluation gaps rather than scaling raw data volume. No concrete benchmark numbers, baseline configurations, or release/access details are disclosed, so the claims remain vendor assertions.
Google has released Gemini 4 Argon, a general-purpose physical reasoning model that achieves an 87% score on the Embodied Reasoning (ERQA) benchmark using pure reasoning. By enhancing spatial understanding, code generation, and visual-semantic reasoning, it supports workflows across the entire robotics lifecycle—from zero-shot control to Real2Sim reconstruction—accelerating progress toward physical AGI.
Google has released Gemini 4 Argon, a general-purpose physical reasoning model for robotics that achieves an 87% score on the Embodied Reasoning (ERQA) benchmark using pure reasoning, outperforming other frontier models. The model demonstrates significant leaps in physical understanding, code generation, and visual-semantic reasoning, supporting applications across the entire robotics lifecycle, including mobile manipulation control and Real2Sim reconstruction of physically calibrated simulation twins from video feeds.
Google has begun rolling out Gemini 4 Argon to paying users, introducing an automated misalignment monitor to prevent malicious use while maintaining previous benchmark capabilities. The model demonstrates significant advances in embodied reasoning, notably identifying critical legacy bugs in Project Wing's drone physics simulator.
Google introduces Gemini Robotics-ER 1.6, significantly improving spatial reasoning and multi-view understanding for robots. The model enhances 'embodied reasoning' to enable greater autonomy in complex environments and fine-grained physical tasks like gauge reading. It is now available for preview on Google AI Studio.