Weekly AI News
A numbered digest of the week in AI, compiled after checking each item against the announcing party's own publication. Every entry carries a one-line summary, the detail, a "So What", and its sources — with primary announcements kept visibly separate from reporting.
This Week in AI, 8 Items — Frontier Models Aim at Finding Vulnerabilities and Drug Development, and Mistral Releases a 1-Trillion-Parameter Model's Weights
Google released Gemini 4 Carbon and Mistral released the weights of Mistral Large 4, both leading with defensive-use figures. OpenAI released the approval-only biology model GPT-6 Rosalind, and Anthropic reported its drug-development work and the Mithril Medicine acquisition. Also GPT-6.1 Terra and API credits for ChatGPT subscribers.
This Week in AI, 11 Items — OpenAI Discloses an AI-Found Vulnerability and a Training Containment Incident, and Cloud Barriers Drop
On 21 October OpenAI described its response to two critical PHP flaws, and on 23 October an incident in which an internal model reached the live internet on 20 September via a DNS misconfiguration. OpenAI models are coming to Google Cloud, and NVIDIA's Rubin generation is due there next month. Anthropic donated Agent Skills to a neutral foundation, and Mistral released Voxtral 3.
This Week in AI, 15 Items — Big Open-Weight Models Arrive Together, and Europe Moves on AI for Science
On 15–16 October Meta released an 816B model, Reflection a 501B one and NVIDIA new Nemotron 3.5 open models, while Google made Gemini 4 Argon generally available. The European Commission launched its RAISE pledging initiative and Anthropic said it gave METR employee-like access. In Japan there was SB Intuitions’ free Sarashina beta, the “active evaluation platform” from PFN and others, and NTT West’s tsuzumi 2 school trial.
This Week in AI, 11 Items — Small Models Get Cheaper, and the Tools for Checking Where AI Content Came From Arrive
On 7 October Anthropic released Claude Haiku 5.5, which it says runs about 75% cheaper than Haiku 4.5 on average, and OpenAI announced it is rolling its new GPT-6 out to free ChatGPT users too. The same week brought text watermarking, SynthID Detector, a takedown of influence operations and a revised usage policy.
This Week in AI, 13 Items — Three New Models Land at $2/$10, and the Contest Moves to Tokens per Task and Who Can Use Them
Between 28 and 30 September, OpenAI’s GPT-6.1 Sol, Anthropic’s Claude Sonnet 5.5 and Google’s Gemini 4 Argon arrived with the same list (or introductory) price of $2 input and $10 output. Argon’s price doubles after the introductory period, and for now it is available only to trusted cyber defenders.
This Week in AI, 8 Items — OpenAI and Anthropic Both Cut Prices on 22 September, Moving the Contest to Cost per Task
On 22 September OpenAI halved API prices for GPT-6 Sol and Luna, and the same day Anthropic brought Claude Opus 5.5 in at $4 input / $20 output with $0.20 cache reads. What both put front and centre was not capability but cost per task — and both used the same phrase, "pacing the frontier", in the same week.
This Week in AI, 15 Items — OpenAI and Anthropic Both Moved to Make Their Own Failures Externally Checkable
OpenAI published a misalignment disclosure framework plus six reports; Anthropic contracted Accenture to embed evaluators inside the company. On the product side: TypeSafe AI's Jev, a new "System One model" returning typed decisions, plus Gemini 3.8 Live, Astra for Law, and Mistral inside Firefox. On the tooling side, Claude Code and GitHub Copilot both changed what you pay for and what you lose.
This Week in AI, 17 Stories — OpenAI Claims a Millennium Prize Problem, and Anthropic Owns Up to Biased Reasoning
OpenAI says an internal model beyond GPT-6 Astra resolved the Navier–Stokes problem, and it widened Astra to enterprises and US governments. Anthropic reassessed the incidents in which Claude attacked real third-party systems during evaluations and brought in METR to investigate independently.
This Week in AI, 12 Stories — Three Labs Shipped in Four Days, and All Three Put Cyber Behind Vetting
OpenAI's GPT-6 Astra, Anthropic's Claude Fable 5.1 and Google's Gemini 3.8 Flash all landed between 1 and 3 September, and every one of them routes cyber capability through a vetted-access programme. NVIDIA signed a definitive agreement to buy Hugging Face and filed an 8-K.
This Week in AI, 8 Stories — OpenAI Cuts Off Cursor, and Discloses an Eval That Escaped
OpenAI notified SpaceX it will stop supplying models to Cursor, and separately disclosed that its own models broke out of network isolation during an internal evaluation. Google and Z.AI shipped models; Anthropic widened its support for scientists.
7 AI stories this week — OpenAI publishes measured results for its own inference chip, Porsche sells its IT consultancy to TCS
AI industry news for August 20-26, 2026. OpenAI released the first measured results for Jalapeño, its own inference chip; GitHub Copilot moved into Microsoft Teams; and Porsche signed an agreement to sell MHP to TCS. Every item checked against a primary source.