New Technology– category –
-
XHEART X7 Ultra: 700 TOPS Autonomous-Driving Chip Specs
BigGo Finance China Overview XHEART has entered mass production of the X7 Ultra autonomous-driving chip, delivering 700 TOPS of compute based on a "large models define chips" architecture. With over 200,000 X7 series units shipped and 10... -
GitHub Copilot Fusion: Model Orchestration vs Claude Code Costs
VentureBeat USA Overview GitHub has introduced Copilot Fusion, an orchestration engine that routes tasks across multiple AI models to reduce costs and latency. Independent testing shows it achieves performance near Claude Opus 5.5 while ... -
LIFT-Bench: AI21 Labs’ framework to curb RL reward hacking
Unite.AI USA Overview AI21 Labs has released LIFT-Bench and LIFT-Train to prevent "reward hacking" in reinforcement learning (RL) by introducing scoped constraints that evaluate both format and content. Testing shows that while tradition... -
Project Maya v1.1.28: DeepSeek Inference Speed and 1M Context
XenoSpectrum Japan Overview Project Maya v1.1.28 improves text generation speeds for DeepSeek models on single-GPU Linux environments, with some RTX 3090 tests showing a 2.2x increase from 12.9 to 27.8 tokens/sec. The update also enables... -
PLaMo 3.1 Prime: PFN’s AI for autonomous research and reports
Impress Watch Japan Overview Preferred Networks (PFN) has released PLaMo 3.1 Prime, a flagship generative AI model featuring a "Deep Research" function for autonomous information gathering and report generation. The model improves halluc... -
Alibaba AI Roadmap: Full-Stack Strategy from Chips to Qwen 5
Philstar.com China Overview Alibaba has detailed a full-stack AI strategy integrating proprietary chips, cloud infrastructure, and the Qwen model series. The company targets a global data center capacity exceeding 20GW by 2032 to support... -
IBM Spyre Accelerator: PyTorch Backend Integration and Specs
Unite.AI USA Overview IBM has integrated its Spyre dataflow AI accelerator as a first-class PyTorch backend as of the second half of 2026. The integration utilizes a rebuilt compiler based on Inductor and native-device registration to op... -
OpenAI Discoveries: Math, Computer Science, and Physics Results
Engadget USA Overview OpenAI has released new AI-generated discoveries across mathematics, computer science, and physics, including a proof challenging the Exponential Time Hypothesis (ETH). The company is collaborating with external exp... -
Clover-35B-A3B-Preview: Diffusion LLM Specs and Performance
PC Watch Japan Overview DiffuSpace has released Clover-35B-A3B-Preview, a diffusion-based language model that achieves generation speeds approximately twice as fast as equivalent autoregressive models in local environments. The model out... -
Liquid AI LFM3: Specs of 2B-class Open-Weight Model Family
Unite.AI USA Overview Liquid AI has released LFM3, a family of six open-weight 2B-class models integrating text, vision, audio, omni, and decision-making capabilities on a common backbone. The LFM3-1.7B text model achieves higher perform...
