Topic
Explore the latest artificial intelligence news and research.
xAI's newly announced Grok 4.5 leads with API pricing that undercuts rival flagship models and higher token efficiency, putting practical cost reduction front and center. Rather than chasing top scores across every benchmark, it focuses on delivering frontier-level performance while keeping total costs down for development and knowledge work.
French company ZML has released a free LLM inference server supporting five major accelerators, including NVIDIA, AMD, and Google. It stands out for unifying the inference environment without locking users into a specific chip, and claims a 10x performance boost for certain models through its proprietary acceleration technology.
In-memory computing, which performs calculations directly within memory, is drawing attention as a way to improve AI power efficiency. A joint US-Korean research team has developed a proprietary memristor SoC with zigzag wiring to eliminate the inefficiency of operations widely used in lightweight AI models, demonstrating extremely high power efficiency that surpasses existing GPUs.
Entire, founded by former GitHub CEO Thomas Dohmke, has unveiled a preview of its decentralized Git network. As massive access from AI agents exposes GitHub's limits, the verifiability of the performance numbers and the involvement of Microsoft-affiliated investors take center stage.
On July 8, 2026, Anthropic announced a new pattern that restricts its top-tier model Fable 5 to an advisory role while delegating execution to Sonnet 5. This article examines the official cookbook's real-world example of costs dropping from $4 to $1.61, along with constraints such as lack of support on Bedrock and Vertex.
Apple has signed a supply agreement with Broadcom worth over $30 billion, extending through 2031 and covering more than 15 billion US-manufactured chips. The major investment aims to secure a stable supply of wireless components and custom ASICs, strengthening America's semiconductor manufacturing base.
On July 8, 2026, OpenAI announced GPT-Live, a full-duplex voice model, and made GPT-5.5-powered voice conversations freely available to free users as well. The enhancement of voice features used by 150 million people weekly could narrow the opportunities for standalone voice AI companies.
Instagui is an open-source tool that uses AI to parse CLI tools' help text and automatically generate a web-based GUI. Its distinguishing feature is that it can turn tools written in a variety of languages into GUIs without requiring developers to modify their code, while balancing ease of use with safety.
Tracing the background of DeepSeek's in-house AI inference chip plan reported by Reuters, and Zhipu AI's search for a custom ASIC. The focus is on the development gap: while Alibaba and Baidu have already reached mass production and IPO stages, both companies remain at an early conceptual stage.
XBM, Intel's proposed new memory concept, is less about replacing HBM4 outright and more about optimizing on-package memory connections using UCIe. The aim is to reduce implementation footprint and boost bandwidth density by leveraging high-speed serial links and placing DRAM within wiring layers.
