DeepSeek publishes technical report for multimodal model
DeepSeek released a multimodal large language model and technical report, proposing an innovative "thinking with…
AI news monthly archive: April 2026 features 272 curated AI news items covering model releases, product updates, industry news, papers, tutorials and opinions, each with an AI summary, source credit and original link.
DeepSeek released a multimodal large language model and technical report, proposing an innovative "thinking with…
The Qwen team released the open-source sparse autoencoder suite Qwen-Scope, turning SAE features into practical tools.…
SoftBank Group plans to launch and push for an IPO in the United States of its newly established AI and robotics…
The Cursor team iterates on and optimizes its agent toolchain the way it builds software products, centered on the…
Anthropic Claude Code team member Thariq Shihipar shared lessons from building the agent around prompt caching: the…
Baidu's ERNIE 5.1 Preview model is officially online. The model adopts a lighter and more efficient architecture,…
Ling-2.6-1T was released last week and officially open-sourced today. The model is positioned as a trillion-scale…
The platform introduced Advanced Account Security features, with core updates including phishing-resistant login…
xAI has launched Custom Voices: about one minute of natural speech is enough to clone a production-ready voice model…
The research team proposed STARFlow-V, a video generation model based on normalizing flows, aiming to address the…
You can absolutely build web applications
The UK AI Security Institute released an evaluation of the cybersecurity capabilities of an early version of OpenAI…
Through a Figma plugin, Codex can now turn implementation plans into visual FigJam whiteboards.
⚙️ We made agent loops run faster with WebSockets in the Responses API. As Codex gets faster, the bottleneck shifts…
Research reveals the propagation path, timeline, and root causes of “goblin” outputs in AI models such as GPT-5. These…
Claude Code includes a built-in skill for operating the Claude Platform. It is suitable for model migration, using API…
A Harvard Business Review article argues that AI's primary economic impact is not automation but the creation of a…
AntLingAGI open-sourced the Ling-2.6-1T model, a new flagship model for real-world agent workflows. As a pioneer among…
The SGLang team (part of LMSYS Org) revealed that the key to fast, efficient, large-scale execution of its flagship…
The article reveals the core mathematical framework underpinning the training and serving of large language models…
OpenAI DevDay is back. San Francisco, September 29
The cost of AI evaluation has crossed a key threshold and is reshaping accessibility. The Holistic Agent Leaderboard…
You can now generate various downloadable files directly in chats with Gemini, including PDFs, @GoogleWorkspace files,…
Manus released Cloud Computer, a dedicated cloud machine that can run bots, Python scripts and scheduled scraping…
Claude Security is now available in public beta to all Claude Enterprise customers. Built on the Claude Opus 4.7…
Google officially released Gemini Embedding 2, a unified embedding model that maps text, images, video, audio and…
The Claude Code team shared core strategies for optimizing prompt caching at scale. Prompt caching works on prefix…
A privacy-preserving analysis of a million conversations shows that about 6% of users seek personal life guidance from…
2025 data shows that AI use at work among US employees has risen from 20% in 2023 to 40%. Companies gaining sustained…
Augment Code added roughly 2.5k characters of Karpathy-style coding rules to AGENTS.md and ran controlled tests with…
The author tested Replit and five alternatives using the same personal finance app prompt. Lovable generated a result…
AntLingAGI has officially open-sourced its trillion-parameter flagship model Ling-2.6-1T. The model uses an…
Claude product manager Jess Yan shared how the beta Claude Managed Agents is changing her workflow. The composable API…
OpenAI is expanding its compute infrastructure project codenamed “Stargate” to provide core compute support for…
Tencent has open-sourced the Hy-MT1.5-1.8B-1.25bit translation model. It has 1.8 billion parameters and is only 440MB…
The Xiaomi MiMo-V2.5-Pro model performed strongly in the latest Arena leaderboard. It ranked sixth globally in Text…
Microsoft and OpenAI ended their exclusive partnership. The next day, Amazon Web Services (AWS) launched three new…
The closely watched trial between Elon Musk and OpenAI CEO Sam Altman opened in federal court in Oakland, centered on…
FlashQLA is a high-performance linear attention kernel developed on TileLang, designed to improve agentic AI…
FlashQLA is a high-performance linear attention kernel built on TileLang, designed for agentic AI on personal devices.…
Page 1 of 7
Sponsored
Orange Relay (pro.nat6.net) is a premium AI API relay — an excellent transfer station with competitive pricing, reliable quality and broad model coverage including Claude, GPT and Gemini.
Visit Orange Relay