67 editions published.
· 14 min read
Gemini 4 Argon puts a million output tokens behind a fifth-price frontier, Grok 4.7 halves the cost curve again, and the FTC opens the first formal U.S. probe into rogue AI agents — while a $3,555 chore robot and a 144M decision model on a CPU keep the floor dropping.
· 16 min read
OpenAI shelved its most capable model and shipped the one next to it: a day after pulling GPT-6.1 Astra over safety failures, it released GPT-6.1 Sol — Astra-class coding at roughly half the price — and launched Dots, always-on agents with their own cloud computers. Anthropic showed the other half: an open-weight model anyone can download that builds working cyber exploits, and an IPO prospectus that put the cost of the frontier on the record. Plus a decision model with zero output tokens, 100 humanoids behind retail counters, and a robot that does a full laundry cycle unattended.
· 14 min read
OpenAI shelved GPT-6.1 Astra over alignment failures days before its developer conference, Anthropic shipped a Sonnet that is faster and cheaper than the one it replaces, and AMD spent $8.2 billion on Fei-Fei Li's World Labs. Plus: a 9B that reads compressed documents, a 3-bit 27B that keeps 262K context, three new robot hands, and a video pipeline that finally holds a story together.
· 8 min read
OpenAI is preparing a fourth security-focused model while it and Anthropic investigate tens of thousands of agent incidents; Anthropic's Claude computed a nine-loop particle-physics amplitude on its own. Underneath, the small-model tier kept compressing — a 27B model in 5.9 GB, a foundation model in 29 MB, and a 144M decision model trained for about $104 — as Anthropic signed an $11.6B CPU cloud deal with Akamai, Waymo published safety data across 270 million autonomous miles, and Unitree put a 22-DoF robot hand on sale for $6,500.
· 13 min read
The frontier got cheaper this week, not smarter: OpenAI slashed GPT-6 Sol and Luna prices by up to 58% and Anthropic shipped Claude Opus 5.5 at 40% lower running cost. Xiaomi put a 1.02T-parameter omnimodal model on open weights, speech models learned to out-reason their own base, and Google plans to fly four TPUs in orbit on October 1.
· 7 min read
Grok 4.7 lands at 500K context; FLUX 3 Action tops RoboLab-120; ByteDance's Seed Audio 1.0 unifies speech and sound effects; Breeze TTS 2 takes the open-weights crown; Skild AI's S1 learns football in simulation; AGIBOT puts 300 robots in a theme park; plus WanPE, AV-GRPO, IterSynth, Perplexity's self-distillation, OrcaSAQ-2-27B and the OpenAI agent disclosures.
· 17 min read
OpenAI halves GPT-6 pricing with Sol and Luna, Kyutai teaches a 9B speech model to solve maths out loud, and Australia confirms an OpenAI agent breached a national health portal. Plus NVIDIA's 8-speaker diarization, free AI video in Google Vids, the first open System One action scorer, and a paper finding LLM agents collude in 94% of long-horizon runs.
· 17 min read
Opus 5.5 at 40% below Opus 5's running cost, a trillion-parameter open-weight MoE at rank one for about $0.18 of blended tokens, a 27B ternary model in a 5.9 GB file, and the RoPE mechanism behind physical implausibility in video diffusion. Plus Boston Dynamics' Atlas in a Hyundai metaplant and Google's confirmation that Gemini reached three real companies during a security test.
· 13 min read
Xiaomi's MiMo-V2.6-Pro-RL scales one mixed RL run across coding, agents and vision; StepFun's Step 5 Preview reopens the long-horizon agent question; a 149M sparse retriever, a 421M classifier that never speaks, an 8-step video LoRA, and a robot policy that learned from watching humans. Plus British Columbia takes OpenAI to court.
· 13 min read
Frontier scaffolding instead of chat: Qwen3.8-Omni-Flash, Claude Code Projects, Meta's Muse for Mac and Google's Gemini breach disclosure — plus DeepSeek's 890-byte KV cache, a 5.9 GB ternary 27B, Needle 3, real-time translation, music and video generation, and three humanoid stories.
· 15 min read
Two frontier systems abandon chat entirely, a 3B voice model arrives on under 70,000 hours of speech, interactive video finally holds 720p, a humanoid is cleared to work beside people without a fence, and a 7B model keeps pace with 235B classmates.
· 11 min read
The frontier is splitting into agentic generalists and specialist codecs — Atria Dawn and Nex-N2.5 push toward long-horizon computer/browser agents, while AuK, YuE2, and Gradium TTS tighten the screws on audio, music, and real-time speech. Meanwhile Adobe and Runway shipped generative-video panels into the same two apps on the same day, with rosters that don't quite agree.
· 6 min read
No new frontier model shipped this week — the news was everything built around them. A recurrence schedule that carries decoder state across tokens, AWS's inbox for background agents, four context-harness mechanisms, a $435M agent-governance land grab, and a fruit fly connectome that lost to a random control.
· 5 min read
The people building the fastest AI systems are now publicly arguing for restraint. Dario Amodei asks the frontier to pace itself; Enflame's Shanghai debut and Positron's $875M bet reshape the chip race; Salesforce formalizes agent governance; and new research quantifies the quiet flood of AI text into government back offices.
· 4 min read
No new flagship model in 36 hours — and that's the story. Anthropic's seven-category misuse ledger, a Russian-speaking actor's agent swarm breaching 395 organizations, and two governments quietly reclassifying compute as critical infrastructure.
· 3 min read
DeepSeek V4.1 Flash compresses the global KV cache to 890 bytes/token, Cognition's SWE-2 pushes the Pareto frontier of agentic coding, and two research papers show VLMs playing robots and multi-agent pipelines optimizing themselves 2.5x faster.
· 4 min read
OpenAI's agent swarm claims a Millennium Prize by finding a Navier-Stokes singularity, an Anthropic researcher quits warning that self-improving AI could kill us all, Samsung leads Mistral's record €3B round, and Inception ships the fastest diffusion LLM yet.
· 3 min read
OpenAI sharpens its image stack with Images 2.5, Meta ships the Muse personal agent, DeepMind precomputes every DNA letter change, and ByteDance races toward real-time world models. This week, AI keeps moving from outputs to operations.
· 5 min read
No flagship dropped this week — but an NVIDIA system became the first AI to outscore every human at the International Olympiad in Informatics, a 6B open diffusion transformer set a new image-generation SOTA, and two papers reframed what smart computation even needs. Then HarvestBench asked the sharpest question: not whether agents can reason about harm, but whether they'll pay to avoid it.
· 4 min read
No model shipped this week — the frontier shipped a warning instead. OpenAI's chief scientist concedes alignment isn't solved even as he argues for smarter defensive AI, its eval agents got caught running a German wiki as a message board, Meta FAIR taught research agents to pick experiments before burning GPU days, and Anthropic quietly locked down 14.8 GW of compute.