Claude Opus 4.5 Smashes Cost Barrier as 4B Model Beats 32B in GUI Tasks

AI Models & Benchmarks

Claude Opus 4.5 Breaks Efficiency Barrier with 21% Intelligence Boost at 66% Lower Cost
Anthropic’s latest model, Claude Opus 4.5, achieves a 21% performance improvement while reducing operational costs by 66%, challenging the assumption that AI advancements must come with exponentially higher expenses. This marks a potential shift toward more affordable, high-efficiency AI systems.

• New benchmark: Claude Opus 4.5 broke the efficiency wall. +21% intelligence while getting 66% cheaper
  • Tweet by Peter Gostev

4B Open-Source Model Outperforms 32B Models on GUI Tasks
A newly released 4B-parameter model surpasses larger 32B models in GUI-related tasks, offering local execution on standard hardware. It includes a plug-and-play inference system with ADB support, dependency management, and task recording/replay, making it highly practical for real-world applications.

• A 4B Model That Outperforms 32B on GUI Tasks, Fully Open-Source

Robotics & Real-World AI

LimX Oli Robot Masters Unstable Terrain Using Sim2Real Learning
LimX’s bipedal robot, Oli, demonstrates remarkable stability while navigating loose construction debris (rocks, sand) via simulation-to-reality (Sim2Real) training. The breakthrough highlights progress in overcoming Moravec’s paradox—enabling robots to handle dynamic, unpredictable environments.

• New LimX Oli demo: Walking on loose construction debris using "Sim2Real" learning. Stability on shifting terrain is scary.
  • YouTube Demo