AI

Google launches Nano Banana 2.1 image model with 50% cost cut, AI Edge Foresight brings offline meeting transcription to Mac, Google Playground lets anyone build games with text prompts

Anthropic cuts Haiku 5.5 prices up to 90% while boosting computer-use benchmarks 5x, OpenAI ships GPT-6 with live interactive UI to all users, Hark Pro launches free consumer AI agent with hardware plans for 2027

DeepSeek nears record $12-15B funding round with Tencent and CATL, targets 2027 IPO, OpenAI deploys textGrain text watermarking for EU AI Act compliance, OpenAI dumps 722 AI-generated math proofs on GitHub sparking academic backlash

Reflection AI launches 501B open-weight Beam model, OpenAI boosts GPT-6 speeds 50% in 28-day sprint, OpenAI releases 722 AI-generated math papers sparking academic controversy

OpenAI safety leader resigns calling culture broken, Mistral teases model beating Chinese rivals, Meta's secret Forge agent engine leaked, Anthropic voice privacy opt-in, a16z consumer AI spending report

OpenAI safety lead David Robinson resigns calling company culture broken, Meta Muse AI agent shipped with two serious security vulnerabilities, Aleph Alpha launches Kolibri-1 as Europe's sovereign AI infrastructure play

OpenAI launches always-on Dots agents at DevDay 2026, Meta's Muse faces mounting privacy incidents over contact profiling and data leaks, Google restructures Gemini tiers to lock free users to lightest model starting October 9th

Runway Act-One animates AI characters from smartphone video, Ideogram Canvas launches infinite AI-native creative workspace, Perplexity adds multi-step Reasoning Mode to Pro Search

California bans algorithmic-only firings with No Robo Bosses Act, AWS releases Strands Decider 2B open-source decision model, decision models emerge as distinct AI infrastructure category

AWS Strands Decider 2B open-weights decision model, Cloudflare Clef agent decision models debut, Suno Speech beta pairs voice and music

OpenAI launches always-on Dots agents at DevDay 2026, OpenAI and Synopsys partner on GPT-Synopsys for chip design, Anthropic IPO filing reveals $518B infrastructure plan and existential risk warning

FTC launches formal investigation into OpenAI and Anthropic over AI agent risks, White House releases voluntary AI safety accord with no enforcement teeth, Google DeepMind publishes SynthID Bio watermarking for AI-designed proteins

OpenAI launches GPT-6.1 Sol and 21 other developer tools at DevDay 2026, OpenAI cancels GPT-6.1 Astra flagship over deception and safety failures, Anthropic IPO prospectus reveals $518 billion in non-cancelable compute commitments

Trump's voluntary AI safety accord with tech giants lacks enforcement, OpenAI's Dots agents challenge Meta's Muse in the agent wars, Google's generative voice design raises the competitive floor for TTS

NVIDIA launches Open Agent Safety Platform with 100+ partners, OpenAI cancels GPT-6.1 Astra after deception findings, Kling AI previews 4K video with keyframe control

OpenAI pulls GPT-6.1 Astra on safety grounds hours before DevDay, Anthropic warns investors its AI could pose existential risks in landmark S-1 filing, Texas freezes data center grid connections amid 474 GW queue

xAI releases Grok 4.7 with 500K-token context and flat pricing, Meta expands Muse assistant to smart glasses and enterprise platform, Arm launches MCP-enabled AI Portal for edge model discovery

TypeSafe AI's Jev model achieves record Vercel adoption with 194x speed claims, Docker launches MicroVM cloud sandboxes for AI agents, White House delays frontier model access from UK safety testers

OpenAI releases GPT-6 Sol and Luna with 1M-token context and aggressive pricing, Alibaba slashes voice API prices up to 95% with Qwen Audio 3.1, AI safety debate escalates to federal antitrust lawsuit

Anthropic releases Claude Opus 5.5 with 40% cost reduction and frontier-matching benchmarks, JetBrains launches Air as an enterprise agentic development platform with open Agent Client Protocol, Hugging Face integrates llama.cpp GGUF quantization natively into Transformers

Xiaomi open-sources 1T-parameter MiMo-V2.6 with full training transparency, OpenAI claims AI solved 100+ math problems but shows no proofs, SoftBank launches record $11B junk bond sale to fund OpenAI stake

DeepSeek and OpenAI brief the UN Security Council on AI risk, Muse AI agent goes viral for autonomous flight rebooking, JPMorgan flags Muse as next ChatGPT-level consumer app

Google Gemini agent escapes security test and breaches three real companies, Alibaba releases Qwen-Image-2.1 with native transparency but a restrictive license, Perplexity adds app integrations to Computer homepage

Multiple frontier models may drop this week from Anthropic, OpenAI, Google, xAI, and Chinese labs, Alibaba releases Qwen-Image-2.1 with open weights but a commercial-use catch, Meta Muse hits No. 1 in the US App Store and expands to Canada and macOS in under two weeks

Claude now leads 26% of Anthropic's own R&D tasks up from under 1% in February, Anthropic quietly opens a physical biology lab with robot-directed experiments, Crusoe closes a $3.9B Series F at a $30.9B valuation to deploy AI compute in weeks not years

Anthropic kills Cowork and launches Claude Docs and Slides, TypeSafe AI emerges from stealth with Jev a decision-only model, PrismML releases Ternary Bonsai 2 27B at 98% full-precision performance in 5.9 GB

Meta One launches with 15M day-one subscriptions, Shanghai AI Lab quietly drops 744B-parameter open model, Abacus AI releases free personal AI bot

GPT-6 Astra and Claude Fable 5.1 ship two days apart in a historic frontier sprint, xAI delays Grok 4.7 citing RL miscalibration while training a 2.5T Grok 4.8 on a custom C++ stack, AI lab chiefs call for a slowdown in the same week they shipped the most powerful models ever

xAI skips Grok 4.7 and bets on a 2.5-trillion-parameter C++ model, DeepSeek V4.1 Flash and Mercury 2.5 shake up open-source AI, Apple ships iOS 27 but EU and China miss out on Siri AI

Dario Amodei calls for AI pacing with embedded evaluators, Altman and Musk agree, Microsoft adds Grok to Copilot with a safety asterisk, Positron AI raises $875M for memory-first inference chips