R1

Reasoning models, represented by OpenAI o1, DeepSeek-R1, and QwQ, have chain-of-thought expansion as their core mechanism, generating a large number of intermediate reasoning steps before producing a final answer.

Rabbit公司的R1设备主打AI代理操作手机应用，未能获得持续的市场认可

Rabbit公司推出的R1设备市场反响平平

The R2 is Rivian's more affordable volume model positioned below the flagship R1 series, and is seen as key to the company's efforts to expand its market share.

In January 2025, DeepSeek released the R1 model, with all weights made public, allowing anyone to use, modify, and commercialize it for free; its performance is comparable to OpenAI O1, and its training cost was only $294,000.

In the inference system description DeepSeek published in February 2025, the cost was calculated as follows: an average of 226.75 nodes running, rented at 2 US dollars per card per hour, for a cost of 87,000 US dollars per day; billed at R1's price at the time, revenue was 562,000 US dollars per day, a cost profit margin of 545%.

In Q3 2023, Rivian was in the midst of a large-scale factory shutdown and renovation, having proactively halted production for several weeks to upgrade the R1 series production platform, which caused production and delivery volumes for that quarter to be significantly depressed.

Rabbit's R1 device marketed itself as an AI-native hardware product, but drew widespread criticism because its core functionality relied on cloud-side app scraping.

Rabbit launched the R1 standalone AI device.

Rivian positions the R2 model as more affordable and lower in environmental footprint compared to the R1 series.

DeepSeek published a paper in the journal Nature, putting R1's final training cost at $294,000 using 512 export-compliant NVIDIA chips.

The Apple Vision Pro features a dual-chip architecture (the M2 handles general computing, and the R1 is dedicated to processing sensor data), ultra-high-resolution Micro-OLED screens (over 11 million pixels per eye), and 12 cameras and 5 sensors.

Vision Pro's hand tracking inference is performed on-device by Apple Silicon chips (M2 and R1), with R1 handling real-time sensor data fusion to compress latency to within 12 milliseconds, and M2 handling skeleton detection neural network inference.

The R1 chip compresses multi-camera latency to within 12 milliseconds, while the M2 chip handles neural network inference for skeleton detection.

The R1 chip handles real-time sensor data fusion, compressing multi-camera latency to within 12 milliseconds, while the M2 chip handles neural network inference for skeleton detection.

Vision Pro is equipped with 12 cameras, and its inference process is completed on-device by Apple Silicon chips (M2+R1).

On Monday, January 27, 2025, following DeepSeek's release of the R1 model, NVIDIA fell approximately 17% in a single day, with its market capitalization evaporating approximately $593 billion, setting a record for the largest single-day loss by any company.