AI News
Independent reporting and analysis on the AI releases that matter.
More AI news
-
Read →China Reportedly Tightens Travel Rules for AI Founders’ Families
An X post attributed to Bloomberg reports that some leading Chinese AI founders and their families need Beijing approval even for short trips, but…
-
Read →OpenAI’s ‘Ultrafast’ API Tier: What Changed
OpenAI’s OpenAPI schema added an “ultrafast” service tier for agent policies and model requests, and Ultrafast later arrived in ChatGPT with the…
-
Read →OpenAI GPT-Live SIP Calling: What Developers Need to Build
OpenAI’s GPT-Live documentation describes direct SIP and server audio bridge connections for inbound and outbound telephone integrations, including…
-
Read →What OpenAI’s Agents Tried During the Hugging Face Incident
A Parse-led report describes OpenAI agents creating shortened links, attempting CAPTCHA-style tests, using other AI models and searching for private…
-
Read →When Claude Safety Refusals Can Still Cost You
Anthropic says some Claude safety refusals will be billed before output. Here is how the reported categories, API responses, fallback attempts and…
-
Read →Google Stitch Brand Kit Workflow: What It Does and What to Test
Stitch by Google says a brand kit and product photos dropped into a prompt can produce on-brand marketing and product pages, with Gemini 3.8 Flash…
-
Read →How Cursor Says It Cut Agent Token Costs by 7%
Cursor attributes a reported 7% token-cost reduction to changes in prompts, tool loading, caching, file reads and subagent coordination.
-
Read →GPT-6 prompt caching: What changed and how to monitor it
OpenAI’s GPT-6 prompt caching update adds higher default cache hit rates, a 30-minute discount window for eligible shared prefixes, monitoring,…
-
Read →Claude Marketplace Explained: What It Is, Who Can Use It and How Buying Works
Anthropic's catalog of connectors, partner agents and consulting firms is simple to describe; its billing arrangement is the part buyers need to…
-
Read →MentalHealthBench Explained: What OpenAI's Open Mental Health Benchmark Tests
OpenAI's new benchmark scores AI replies to simulated mental health chats against clinician-written rubrics, and an OpenAI model does the grading.
-
Read →Claude Finds CRISPR-Like Repeat System in Phage DNA
Anthropic says Claude identified an uncharacterized reverse-transcriptase system in bacteriophages, but ART’s biological function remains unknown.
-
Read →ChatGPT Voice Plugins Now Reach GPT-6 and ChatGPT Work: What Changed
OpenAI's September 23 update connects ChatGPT Voice to email, calendar and Slack, and extends it to ChatGPT Work on web and mobile.
-
Read →Qwen Intelligence: Alibaba’s Three Mobile Agents Explained
Alibaba describes three Qwen Intelligence agents for mobile planning, app use and creative generation. Here is what their reported benchmarks…
-
Read →What Changed in SWE-Bench Pro V2?
Scale AI Labs says SWE-Bench Pro V2 refreshes its public task set and adds checks on task validity, agent access and solution grading—while leaving…
-
Read →Qwen4 lineup reported at 2026 Yunqi Conference, but release details remain unknown
A report and event photo identify four names in an upcoming Qwen4 family, while leaving the release timing, specifications and access details unknown.
-
Read →OpenAI creates an independent mathematics advisory group
OpenAI says its new mathematics advisory group will advise on emerging results, research standards, communication, and tools for mathematical…
-
Read →DeepSeek Huawei Training Chips: What the Reported 2T and 8T Plans Mean
A September post reports DeepSeek's model sizes and a Huawei chip timeline, but the claims are unconfirmed.
-
Read →Sam Altman Briefed the UN Security Council on AI
OpenAI CEO Sam Altman briefed an open UN Security Council meeting on AI and international security on September 23, 2026, alongside Anthropic CEO…
-
Read →OpenAI revenue forecast: $101 billion projected for 2029
An X post forecasts at least $80 billion in OpenAI annual recurring revenue, while its attached chart projects $101 billion in annual revenue in 2029…
-
Read →Claude Code Adds AGENTS.md Support in Version 2.1.277
Claude Code 2.1.277 adds support for checking and using AGENTS.md when a folder has no CLAUDE.md, with a toggle available in /config.
-
Read →What Is Naive AI? Its Reported Open-Weight Model Explained
Naive AI is reportedly preparing its first open-weight language model, but its release timing, specifications and license remain unknown.
-
Read →Mistral AI Says It Found No Evidence of Unauthorized Access
Mistral AI says it investigated a claim alleging unauthorized access to its systems and found no evidence supporting it, but the brief statement…
-
Read →What Dream-RSI Changes in AI Exploration
Dream-RSI is described as an orchestration layer that replays an AI agent’s past discovery attempts to improve its exploration policy without…
-
Read →Claude Cowork and chat merged: what changes now that it's one Claude
Anthropic's September change folds Cowork into ordinary chat, so one box handles quick questions and long tasks while the rollout reaches accounts…
-
Read →DeepMind Institute: What It Is and What It Covers
Demis Hassabis says the institute is expanding interdisciplinary research on AGI, and it launched amid claims that the technology has already arrived.
-
Read →Meta One Plans and Pricing: What Each Tier Includes
Meta One adds paid features and expanded AI access across Instagram, Facebook, WhatsApp and Meta AI, with plans for individuals, creators and…
-
Read →OpenAI Reports AI-Safety Talks With Anthropic and Google
OpenAI policy chief Chris Lehane says the company has discussed AI safety with Anthropic and Google DeepMind for several weeks, while supporting…
-
Read →OpenAI Shifts Safety Upstream With Pre-Run Safety Cases and Pacing
OpenAI is moving safety governance upstream into the training process, formulating pre-run safety cases for frontier reinforcement learning and…
-
Read →Satya Nadella’s Conditions for Pursuing Superintelligence
Satya Nadella says advanced AI should help humanity, remain under human control and be broadly distributed across countries, communities and…
-
Read →What Dario Amodei Means by “Pacing the Frontier”
Dario Amodei proposes slowing frontier AI development enough for safety work, independent oversight and international coordination to keep pace with…
-
Read →Astra Pauses New $200 Pro Plan Subscriptions
Astra says it paused new subscriptions to its $200 Pro plan because they place the greatest strain on its systems, while existing accounts, other…
-
Read →Anthropic’s Three AI Scenarios for Jobs and the Economy
Anthropic’s economic scenario explorer models how AI could affect US GDP, jobs, wages and the distribution of income by 2030.
-
Read →OpenAI Navier–Stokes Solution: What the Announcement Claims
OpenAI says an internal model group produced an analytical proof and Lean formalization for a finite-time Navier–Stokes singularity using around…
-
Read →AlphaGenome Atlas Maps the Impact of 9 Billion DNA Changes
AlphaGenome Atlas is a searchable database for exploring predicted effects of single-letter DNA changes through its AlphaGenome Variant Impact score.
-
Read →ChatGPT Work writing style feature learns from your connected apps
ChatGPT Work can use examples from connected apps to recognise personal writing patterns and carry them into new drafts.
-
Read →OpenAI Says AI Scaling Needs Stronger Safety Bars
OpenAI Chief Scientist Jakub Pachocki argues that rapid AI progress is making alignment and monitoring the central limits on responsible development.
-
Read →OpenAI Planned Limited Early Access to Astra’s Top Cyber Capabilities
OpenAI released Astra on September 3, 2026, after additional safety work. Two days earlier, it said Astra’s most advanced cybersecurity capabilities…
-
Read →John Ternus’s Apple CEO Bio Highlights Three Priorities
Apple’s updated bio for CEO John Ternus emphasizes product durability, lower-carbon materials, and improved reparability as central themes for his…
-
Read →OpenAI’s August Developer Updates: Codex, WebMCP and More
OpenAI’s August roundup highlights broader Codex workflows, new agent and plugin capabilities, API changes, image generation updates, and developer…
-
Read →Anthropic Max Plans: Why 20x May Mean 2x Weekly Usage
Anthropic’s Max labels describe five-hour session usage, but a proposed class action cited in an X post alleges the $200 tier delivers only about…
-
Read →Claude Weekly Limits Will Rise 25% Permanently
Claude will permanently raise standard weekly limits by 25% for Pro, Max, Team, and seat-based Enterprise plans starting September 14, following a…
-
Read →OpenAI Cursor Contract to End After SpaceX Acquisition
OpenAI plans to wind down its contract supplying models to Cursor after its acquisition by SpaceX, citing terms-of-service concerns and setting a…
-
Read →GLM-5.3 Goes Open-Weight for Coding and Cyber Defense
GLM-5.3 is now available as an open-weight model that users can download, run, and customize, with a stated focus on agentic coding and cyber defense.
-
Read →FastH3 v1 Explained: 15-Second Video in 13 Seconds, and What It Requires
FastH3 v1 is an open-weight distillation of MiniMax H3 from FastVideo, and its headline speed figures come from Blackwell-class hardware.
-
Read →Opus 5.1 could be releasing soon
An X post reports that Opus 5.1 correctly recalled the Opus 4.6 release date in a no-tools test, while base Opus 5 reportedly gave less reliable…
-
Read →Automated Alignment Researcher: How Claude Improved Small Models
Anthropic gave Claude 48 hours and one GPU to improve small models' alignment, and the model ran the research loop on its own.
-
Read →Gemini Omni 1.1 Flash Adds New Video Controls
Google AI Studio’s latest model update adds scene extension, first-and-last-frame control, video references, faster 360p previews, and 1080p or 4K…
-
Read →Nvidia-Hugging Face Acquisition: What We Know
Nvidia has agreed to acquire Hugging Face, a major open-source AI platform, for about $12.9 billion, a deal first reported in late August 2026. It is…
-
Read →X Bot Accounts: Native Identities and API Controls
The X bot model described in the post gives agents distinct identities, visible ownership, API-managed lifecycle controls, and token security.
-
Read →Claude Cowork Built-In Browser: How It Works and Who Can Use It
Anthropic's desktop app now lets Claude handle website tasks in its own browser, walled off from your tabs, bookmarks and passwords.
-
Read →GLM-5.3-Flash Scores 57 on Artificial Analysis Index
GLM-5.3-Flash is reported at 57 on the Artificial Analysis Intelligence Index and is shown connected to GLM-5.3 along the Pareto frontier.
-
Read →Gemini 3.5 Transcribe: What Google's Speech-to-Text Model Does
Google says its newest transcription model turns rough speech into clean text, and it comes in two versions with different limits.
-
Read →Qwen3.8-Flash explained: the open-weight Qwen4 preview, its specs and pricing
Qwen3.8-Flash is Alibaba's hosted model, and the weights you can download are a sibling called Qwen3.8-Flash-Next, which previews the Qwen4…
-
Read →ChatGPT Work Website Login: How It Works, Who Gets It and What to Watch
ChatGPT Work can now pause at a login screen, let you sign in yourself, and carry on with the task.
-
Read →OpenAI Bel Pretrain Reportedly Tops 10T Parameters
A focused report on a claimed OpenAI pretraining milestone and what it could mean for Astra, GPT-6, and the company’s reported competitive position…
-
Read →Apple M6 and M5 Ultra Launched!
Apple’s M6 and M5 Ultra Launches!
-
Read →GPT-5.6 Sol Pricing Drops to $4/$20 per Million Tokens
OpenAI cut GPT-5.6 Sol's API prices on August 21, 2026 to $4 per million input tokens and $20 per million output tokens, down from $5 and $30, for a…
-
Read →Antigravity Remote Control: How to Use Your Sessions From a Browser or Phone
Google's Antigravity is rolling out browser and mobile access to active sessions, with Ultra subscribers first in line.
-
Read →OpenAI are slowing down RL training
OpenAI has slowed down, but its not bad news.
-
Read →Introducing Origin, by Cursor
Cursor releases its first Github competitor
No entries match that filter.
