Perceptron Releases Mk1.5 Embodied Multimodal Model

// 1h agoMODEL RELEASE

Perceptron Releases Mk1.5 Embodied Multimodal Model

Perceptron has unveiled Mk1.5, an upgraded multimodal foundation model engineered for real-time physical AI applications across robotics, drones, and smart wearables. The model integrates vision, audio, and language reasoning with native spatial output primitives and low inference latency for real-time control loops.

// ANALYSIS

While mainstream LLMs prioritize text and digital workflows, true physical autonomy demands grounded spatial reasoning at interactive speeds, positioning Perceptron Mk1.5 as a pivotal bridge between high-level reasoning and physical actuation.

–Dramatic latency improvements (up to 4.7x speedup over Mk1) make multi-billion parameter foundation models viable for dynamic drone and robotic control loops.
–Native structured outputs (polygons, bounding boxes, tracking points) eliminate the friction and latency of chaining disjointed computer vision pipelines.
–Training on first-person egocentric video tailors the model directly for wearables, autonomous inspection systems, and quadrupedal robots.
–Broad availability via OpenRouter democratizes access to embodied physical intelligence without requiring massive localized compute clusters.

// TAGS

perceptronperceptron-mk1.5embodied-airoboticsmultimodalcomputer-visionphysical-aifoundation-models

DISCOVERED

1h ago

2026-09-25

PUBLISHED

1h ago

2026-09-25

RELEVANCE

8/ 10

AUTHOR

shiparena

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

NEWS31m ago

Macroscope Post-Trains Code Models on Fireworks AI

Building on its MacroscopeBench benchmark, software intelligence startup Macroscope announced it is post-training in-house code-review models using Fireworks AI's infrastructure. Trained strictly on open-source software defect datasets rather than customer code, the custom models cut agent reasoning steps by 25% while maintaining evaluation performance.

RESEARCH42m ago

Salesforce SFR-AutoR&D Autonomously Invents Machine Learning Methods

Salesforce AI Research developed SFR-AutoR&D, an autonomous agent system engineered to formulate, code, and execute machine learning methodologies end-to-end. Algorithmic techniques devised by the agents achieved up to +14 point benchmark gains and a 3.14× generation speedup in reinforcement learning, while infrastructure optimizations enabled 1-trillion-parameter model LoRA fine-tuning on a single 8-GPU H200 node.

UPDATE1h ago

Claude Opus Adds Helicopters to SomethingBig World

Matt Shumer showcased an update to Something Big World, a browser-based 3D virtual environment where an autonomous agent powered by Claude Opus 5.5 is running in an iterative development loop. The model independently decided to add helicopters to the virtual city simulation and stated it would continuously refine their implementation and mechanics over the course of the day without direct human steering.