What Is Reasoning
A new paper and online discussions have revealed methods to extract hidden reasoning traces from closed-weight AI models, prompting an investigation into how these traces work. Reasoning traces are si…
DeepSeek is a Chinese AI research laboratory that has developed highly capable open-source language models including DeepSeek-V3 and DeepSeek-R1, notable for their efficiency and performance.
A new paper and online discussions have revealed methods to extract hidden reasoning traces from closed-weight AI models, prompting an investigation into how these traces work. Reasoning traces are si…
Maritime, a new platform, offers isolated AI agent hosting for $1 per agent per month, with a free tier for up to three agents, aiming to eliminate infrastructure complexity for developers. The servic…
Nvidia and more than 200 companies signed an 'Open Weights and American AI Leadership' letter in July, arguing that open-weight AI models can accelerate innovation, but the industry must solve the fun…
ChatOSS, a native desktop workspace for agentic coding built on Ollama, launched as an open-source alternative to Codex, offering AI agents that work inside repositories with an integrated Kanban boar…
Enterprises should prioritize building datasets and customizing small AI models over relying solely on cloud-based frontier models, according to a tech commentator. The argument emphasizes that custom…
World Liberty Financial, backed by President Donald Trump, is collaborating with Hong Kong-based AI company WorldClaw, which offers 43 Chinese-developed AI models out of 90, including from Alibaba, Ba…
DeepSeek released DeepSeek V4 Pro 0813, an updated version of its V4 Pro model, featuring the same core architecture with approximately 1.6 trillion total and 49 billion active parameters, a new DSpar…
Most AI models marketed as 'open' are open-weight rather than open-source, and the deciding factor is the licence, according to the Open Source Initiative's OSAID 1.0 standard. Open-weight models like…
Chinese open-weight AI models from DeepSeek, Alibaba's Qwen, and Moonshot AI's Kimi now account for a majority of global AI workloads after frontier-grade intelligence costs collapsed by roughly 90 to…
Brokk AI released Muninn, a 346M-parameter code localization model that outperforms OpenAI's text-embedding-3, and Muninn-small (47M parameters) that runs on CPU, both Apache-licensed and integrated i…
A developer outlines how startups and enterprises should choose AI API stacks, arguing that direct provider access often fails both groups due to payment barriers, pricing, and support gaps. The piece…
World Liberty Financial told CoinDesk it does not own or operate AI aggregator WorldClaw and has no control over which AI models the platform offers, but declined to detail any financial relationship.…
The KV cache has no standard ABI, with vLLM's FlashAttention backend alone reporting its cache shape as a four-dimensional tensor that varies by backend, attention variant, and model family, complicat…
Krystal Loop Protocol (KLP) is a new operating pattern for AI coding agents that aims to keep multi-agent projects coherent and under human control by using a bounded worker/critic loop. The protocol,…
A developer benchmarked ten AI coding models across five tasks and four languages to find the best value for production use. The results show DeepSeek V4 Flash and DeepSeek Coder offer the highest val…
Alibaba's Qwen3.8-27B, a 27-billion-parameter model, matched OpenAI's GPT-5.6 Luna on the Artificial Analysis Intelligence Index, outperforming larger rivals while running on everyday hardware. The mo…
VIDRAFT, a Korean deep-tech AI/science research company, has launched the Open Discovery Challenge, a public drug discovery competition hosted on Hugging Face that evaluates AI-generated molecular can…
Anthropic's annual revenue run rate has reached $65 billion ahead of its expected IPO later this year, seven times its $9 billion sales at the end of 2025, according to Bloomberg. The company plans to…
A developer with a background in banking, food production, and construction built Jimi, a personal AI assistant for Telegram, using a FastAPI backend decoupled from the messaging layer. The assistant …
Microsoft Corp. has quietly wound down its five-year retreat from China, shuttering offices, closing retail stores, and cutting thousands of jobs, while continuing to sell Azure and AI services to Chi…