LlamaIndex open-sources LiteParse local document parsing tool for AI agents
LlamaIndex open-sourced LiteParse, a fully local CLI and native TS library that extracts layout-aware text from PDFs,…
AI news monthly archive: March 2026 features 171 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.
LlamaIndex open-sourced LiteParse, a fully local CLI and native TS library that extracts layout-aware text from PDFs,…
Cognition announced that Devin can break large tasks into parts and delegate them to multiple managed Devins running…
A new toolkit covering six protocols, including MCP and A2A, has been released to eliminate the need for custom…
PromptArmor summarized its research on prompt injection and data exfiltration in spreadsheet AI features, covering…
(1/N) We are launching Dreamverse. Most AI video models take minutes to generate a 5-second 1080p clip. In 4.5…
The FastVideo team released the Dreamverse prototype interface, introducing an innovative vibe-director workflow. This…
OpenAI released GPT-5.4 mini, now available in ChatGPT, Codex, and the API. Optimized for coding, computer use,…
GPT-5.4 mini is available today in ChatGPT, Codex, and the API. Optimized for coding, computer use, multimodal…
Hugging Face published a blog post about the open-source ecosystem on its platform. Written and published by Hugging…
M2.7 is the first model in the M2 series to deeply participate in its own evolution. It can build complex agent…
LangChain released the open-source framework Open SWE, distilling common architectural patterns from the internal…
Company H released Holotron-12B, a multimodal computer-use model. Built on NVIDIA's open-source Nemotron-Nano-12B-VL…
GPT-5.4 mini and nano were released as lightweight, high-speed versions of GPT-5.4, optimized for coding, tool…
Codex added Subagents, enabling the creation of specialized subagents to handle tasks in parallel, keep the main…
Although GPT 5.4 is strong, its most prominent trait is human-likeness. Compared with 5.3, which is already good at…
In its first week live, the GPT-5.4 API processed 5T tokens per day, with traffic exceeding the total volume of the…
Sam Altman praised the Codex team as hardcore builders with a solid product, noting that hardcore developers in the…
Within one week of launch, GPT-5.4 reached 5T tokens processed per day, exceeding the entire API's total volume in the…
Manus released a major upgrade to its Google Workspace connector, integrating the Google Workspace CLI open-sourced on…
OneClaw, an open-source project by an individual developer supported by Kimi, has surpassed 100,000 downloads. It…
OpenAI CEO Sam Altman acknowledged that scaling model size alone cannot reach AGI and that major architectural…
The UK AI Security Institute published a blog post on evaluating the multi-step cyberattack capabilities of seven…
Mistral released Mistral Small 4, unifying Magistral's reasoning, Pixtral's multimodal and Devstral's agentic coding…
Manus released "My Computer," the core feature of the Manus 2.0 desktop app, extending agents from the cloud to users'…
Claude announces that for the next two weeks, usage limits will double on weekends and outside 5-11 PM Pacific Time on…
Claude Code launches remote control: after running `claude remote-control`, users can start new sessions on their…
YC-incubated Compresr released Context Gateway, which automatically compresses overly long conversation history…
Claude Code enables the Opus 4.6 1M model by default for Max, Team and Enterprise subscribers, supporting a 1 million…
The 1M token context window for Claude Opus 4.6 and Claude Sonnet 4.6 is now generally available (GA), allowing users…
Dylan Patel breaks down the three core bottlenecks constraining AI compute expansion: power infrastructure limits,…
The 1M token context window for Claude Opus 4.6 and Sonnet 4.6 is now generally available and billed at standard…
Claude added interactive chart and diagram generation, supporting the creation of visual content directly in chat. The…
A survey of the core realities and key characteristics of the current stage, analyzing important changes and future…
After analyzing PyPI data, Answer.AI found that the Python package creation rate did not rise noticeably after…
Claude for Excel and Claude for PowerPoint added cross-file context synchronization. When multiple files are open at…
It is not that the IDE era is over, but that we need a "bigger IDE" capable of managing AI agents. The basic unit of…
Claude launched a beta visualization feature that generates interactive charts, diagrams and other visual content in…
LlamaIndex published an article explaining the root cause of why AI agents struggle to read PDFs: PDFs originate from…
How ChatGPT defends against prompt injection and social engineering: constraining high-risk actions and protecting…
OpenAI built an agent runtime on the Responses API, shell tool and managed containers, supporting file handling, tool…
Page 3 of 5
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