AI News: 1x Neo Robot, Extropic TSU, Minimax M2, Cursor 2, and more!

The latest AI advancements include 1X's humanoid robot Neo for home use, Extropic's energy-efficient thermodynamic computing, and new powerful open-source models from Minimax and IBM.

• 1X's Neo is the first mass-market humanoid robot for homes, pre-orderable and comparatively affordable.

• Extropic introduces Thermodynamic Sampling Units (TSUs) for computing, promising significantly higher energy efficiency.

• New open-source AI models like Minimax M2 and IBM Granite 4.0 Nano push performance and efficiency boundaries.

1X has launched Neo, the first pre-orderable mass-market humanoid robot for home use, priced similarly to inexpensive cars and capable of lifting 150 pounds, although initial demos suggest it often requires tele-operation. Extropic is pioneering thermodynamic sampling units (TSU) for compute, claiming up to 10,000 times greater energy efficiency than traditional CPUs/GPUs by sampling from probability distributions, a method foundational to AI. Major AI model releases include Minimax M2, an open-source Chinese model outperforming Gemini 2.5 Pro, and IBM's Granite 4.0 Nano, a family of small, efficient models designed for edge applications.

1X Neo Robot Launch

• 00:00:05 1X has introduced Neo, a humanoid robot available for pre-sale for home use, marking the first mass-market pre-orderable device of its kind. Priced at $20,000 or $499/month, it is comparable to the cheapest cars in the US, weighs 66 pounds, can lift 150 pounds, has 22 degrees of hand movement, and operates quietly at 22 decibels. While envisioned for tasks like laundry and tidying, initial demos indicate significant reliance on tele-operation, raising questions about its autonomy at its anticipated 2026 launch.

Extropic's TSU Innovation

• 00:02:13 Extropic has unveiled its Thermodynamic Sampling Unit (TSU) hardware, representing a fundamentally new and highly energy-efficient computing paradigm. Unlike traditional CPUs or GPUs that execute deterministic commands, the TSU samples from probability distributions, mirroring how AI systems operate. Extropic claims this novel hardware, which also requires new software, can achieve up to 10,000 times greater efficiency than existing computing platforms, though it remains in the prototype and simulation phase.

New Open-Source AI Models

• 00:04:03 Two significant open-source AI models have been released: Minimax's M2 from China and IBM's Granite 4.0 Nano. Minimax M2 has achieved a new intelligence record for open-weights models, scoring 61 on the artificial analysis index, placing it above Gemini 2.5 Pro with only 10 billion active parameters out of 200 billion total. IBM's Granite 4.0 Nano models, including 1.5 billion and 350 million parameter versions, are designed for edge and on-device applications, demonstrating excellent performance for their small size and commitment to efficient enterprise solutions.

Tesla's Distributed Inference Fleet

• 00:08:22 Elon Musk has proposed leveraging the idle supercomputing hardware in Tesla vehicles to create a massive distributed inference fleet. Since Teslas are equipped with high-end hardware for autonomous driving, this compute capacity largely sits unused when cars are parked or charging. Musk envisions that with tens of millions of cars, this could amount to 100 gigawatts of distributed inference capability, potentially allowing Tesla owners to earn money by contributing their vehicle's idle processing power to AI tasks.