Last Week in AI

Big model moves and pricing wars dominate the news, from a major new default model with surprising safety flags to aggressive upgrades and voice tools from rival builders. A quirky training bug turned models into goblin-obsessed storytellers. Corporate and government drama heats up with courtroom turns, SpaceX compute deals, Pentagon AI deployments, and expanded pre-release model reviews.

A rapid tour of this week’s big AI moves: a new GPT-5.5 release with coding gains and strange 'goblin' quirks. Real-time voice agents claiming benchmark dominance and Starlink automation wins. DeepSeek V4’s MoE scaling and 1M‑token context via compressed attention. Major deals and legal drama across Google, Anthropic, Meta, and OpenAI. New safety research on sabotage, document degradation, and bit‑flip attacks.

They review breakthroughs in image models that generate crisp, readable text and screenshot-like results. They cover major Chinese and open model releases, including a trillion-parameter MoE and new API-only offerings. They discuss research tools that automate deep investigations and how AI is being used to find and fix hundreds of software bugs. They also highlight big cloud investments, chip and startup moves, and rising AI-generated media spam.

Coverage of major model releases like Opus 4.7 and Muse Spark, including new reasoning controls, vision and memory upgrades, and evaluation-awareness findings. Discussion of GPT-5.4-Cyber and Codex enhancements for security, plugins, and long-horizon tasks. Rundown of agent products, compute deals, legal battles, and geopolitical risks to AI infrastructure. Exploration of multi-modal 3D world models and AI-driven propaganda trends.

A rapid roundup of Anthropic’s Project Glasswing and the preview of Claude Mythos with troubling cyber and bio capabilities. Coverage of Google’s Gemini 3.1 real-time multilingual voice and Z.ai’s open-weight GLM-5.1 multimodal release. Business moves like Anthropic’s huge revenue run rate, SoftBank’s loan for OpenAI, and major acquisitions and valuations. Research highlights include LLM emotion concepts and training metagaming.

They unpack OpenAI pulling a video app and pivoting toward coding and productivity agents. They cover agents gaining full desktop control and phones running background task automation. New coding and image models, chip and memory industry moves, and robotaxi scaling get quick attention. They flag policy proposals, monitoring and shutdown-resistance research, experiments on model preferences, and the rise of self-improving hyperagents.

Breaking AI product launches and model releases are unpacked, from ultra-long-context mini/nano models to an open-sourced MoE family. Local and sandboxed agent runtimes for Macs and servers are debated alongside Nvidia’s graphics AI and chip forecasts. Business pivots, geopolitical GPU moves, and fresh safety and research work on steganography, chain-of-thought, attention residuals, and state-space model advances are highlighted.

Big AI model and hardware reveals, including a new hybrid 120B parameter model and shifts in GPU supply. Local and always-on coding agents get attention alongside in-chat interactive visuals. Legal and business upheaval surfaces with a major company lawsuit and high-profile departures. A wave of research explores model steering resistance, long-context memory tricks, RL for code generation, and multimodal pretraining.

Big model releases take center stage: a new GPT with a 1M-token context window and mid-response correction, and a faster Gemini Flash Lite with Gmail/Drive/Docs agent tools. A multimodal player unveils unified models for rapid creative work. Industry drama heats up around defense contracts, supply-chain risk labels, leadership departures, a high-dollar funding round, and a controversial lawsuit.

They cover Anthropic’s Sonnet 4.6 with 1M context and ARC-AGI-2 gains. Google’s Gemini 3.1 Pro and xAI’s Grok 4.2 beta get attention for multimodal demos and multi-agent debate. Big funding and compute moves include Meta–AMD talks, MatX and World Labs raises, and a startup simulating human behavior. Research highlights include deep-thinking tokens, optimizer tricks, attractor states, and Anthropic–Pentagon tensions.