* Performance: The model is optimized for “agentic coding” tasks. In benchmarks, it scored 59.5 on SWE-bench Pro, surpassing Google’s Gemini 3.1 Pro and slightly exceeding OpenAI’s GPT-5.5. It also performed strongly on other agent and reasoning tests.

* Inference and Release: Before its official launch, it operated anonymously on OpenRouter as “Owl Alpha,” becoming one of the platform’s top three most-used models. The model weights and technical infrastructure are expected to be released soon on platforms like Hugging Face. API pricing is set at $0.75 per million input tokens and $3 per million output tokens, with promotional rates available.

---

Meituan trained LongCat-2.0 on over 50,000 unnamed Chinese AI ASICs arranged in superpods with high-bandwidth interconnects. The chips share architectural similarities with Huawei’s Ascend 910C series, though Meituan has not publicly named the exact vendor.

The training run consumed more than 35 trillion tokens, including hundreds of billions of tokens with approximately 1-million-token context lengths. This level of scale — previously achieved only on NVIDIA GPUs or Google TPUs — required extensive custom engineering in parallelism, fault tolerance, and numerical stability.

The team implemented 6D parallelism (tensor, context, expert, data, pipeline, and embedding parallelism) to efficiently distribute both the MoE layers and the novel embedding components across the cluster.

Scientists at Delft University of Technology, in collaboration with Waymo, have developed a new model that predicts with high accuracy how human drivers respond to dangerous traffic situations. For the first time, different types of collision avoidance behavior are combined into a single model. The results will be published on 10 June in Nature Communications. Waymo is already using the model to compare the performance of its autonomous vehicles with that of human drivers.

When a leading vehicle suddenly brakes or an oncoming car unexpectedly enters your lane, you have only fractions of a second to decide whether to brake, swerve or both. “Existing models typically describe only part of this process, such as reaction time or steering behavior,” says Arkady Zgonnikov, assistant professor at Delft University of Technology (The Netherlands). “Our new model brings all these components together.”

The model integrates perception, decision-making and execution into a single coherent framework. As a result, it can detect when a situation becomes dangerous, predict how the traffic situation is likely to evolve and simultaneously determine the most effective avoidance strategy.

The first global safety regulations for fully autonomous vehicles were adopted at the U.N., setting uniform international requirements that could pave the way for larger-scale rollouts. The framework is expected to enter into force in January 2027.

---

The first global regulations for fully autonomous vehicles were adopted Wednesday, a U.N. agency said, establishing uniform international safety requirements that could pave the way for larger-scale rollouts of self-driving cars.

Safety concerns and the cost of developing next-level systems have long slowed progress on autonomous vehicles.

As self-driving cars have begun to hit the road in a growing number of cities, fragmented national approaches to regulation have spurred manufacturer fears that vehicles developed for one market could be blocked from others.