Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
VLLM, a model-serving engine for large language models, introduced a small op-level IR to resolve the tension between acting as a compiler target and a hand-tuned kernel dispatcher. The IR allows vLLM to lower operations…
Google researchers found that inference compute budget fundamentally changes how frontier LLMs should be evaluated. Greg Brockman signaled GPT-Realtime-2 is a meaningfully distinct new model. Anthropic reported that 80% …
FusionHarness, an open-source mixture-of-agents compound-model server, was released as a self-hostable alternative to OpenRouter's Fusion API. It fans out prompts to a panel of LLMs in parallel, uses a judge to extract s…
AI coding agent extensions can degrade performance when installed together due to token competition, vocabulary collisions, and guidance conflicts, even if each extension works well in isolation. Developers face unpredic…
Google debuted Infinite Scaler at Google I/O 2026, a game where attendees generated 3D levels in real time using text prompts and reference images via the Gemini API. The game showcased Gemini's speed at 1,500 tokens per…
A developer tested three AI documentation tools—Mintlify, Document360, and GitBook—on a production SaaS codebase with 12K lines of code and 40+ endpoints. Mintlify achieved 94% auto-doc accuracy with 42-minute setup, Doc…
Pramaana Labs, an AI startup founded by Ranjan Rajagopalan, raised $27 million in seed funding led by Khosla Ventures to develop formal verification systems for large language models. The company targets rule-heavy domai…
On May 6, 2026, Anthropic launched Dreams for its Managed Agents platform, followed by OpenAI's Dreaming V3 for ChatGPT on June 2, 2026. Both features use background processes to consolidate memory from past sessions, bu…
Whire's EU payment agent architecture uses a consent_url pattern where an LLM drafts payments and a human approves before execution, solving hallucination risks for low-volume transactions. However, as scale increases to…
A developer built their first Model Context Protocol (MCP) server to give Claude Code read access to a local project knowledge base, sharing four lessons about tool design. The server uses the official @modelcontextproto…
OpenAI researchers developed a method called Deployment Simulation that predicts AI model failure rates before release by using real, anonymized user conversations instead of synthetic tests. In tests with GPT-5 models, …
Google announced its first new smart speaker in six years, the $99 Google Home Speaker, which ships June 25. The device is built around Gemini for Home, but key features like Gemini Live and Home Briefs require a $10/mon…
Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman are attending a G7 CEO lunch in Evian-les-Bains, days after the U.S. blocked EU access to Anthropic's latest AI models. European leaders aim to use the meeting to rebu…
A developer released Superpowers, an open-source framework that gives Claude Code a structured workflow: plan before acting, test before asserting, and prove completion. The plugin, available via Anthropic's official mar…
Apple's iOS 27 update introduces four new features to the Notes app, including Siri AI integration for generating notes, Markdown copy-and-paste, divider lines, and section links. The update is currently in developer bet…
A user tested Claude's ability to generate Lean 4 code to formalize a ring theorem, achieving a proof after 11 iterations but with five unproven 'sorry' sections. The experiment highlights challenges in AI-assisted forma…
A developer explores when it is acceptable to forward AI-generated output to another human reader, while another engineer outlines three methods for AI coding at home without enterprise-level spending. Daniel Stenberg ad…
A developer built a Chrome extension that uses Claude AI to detect phishing messages. The extension sends suspicious messages to a Cloudflare Worker, which proxies requests to the Anthropic API using the Haiku model for …
Pramaana Labs raised $27 million in seed funding led by Khosla Ventures to apply formal verification to AI systems, targeting high-stakes fields like law, drug discovery, and tax preparation. The startup combines large l…
A developer built a budget model comparing the real cost of Claude Code, OpenRouter, and image APIs, revealing that production costs often exceed listed prices due to retries, planning buffers, and failed jobs. For a sam…