OpenAI sharply expanded GPT-6.1 Sol's default context to 500k tokens in Codex and Work (up ~80% from 272k) and stopped charging long-context fees for subscription usage — a major practical upgrade for long-running coding agents. The AI-mathematics upheaval continued, with mathematicians voicing shock and frustration at OpenAI's mass proof releases even as "cyborg" human-AI research workflows start to dominate the field.
The AI-welfare debate went mainstream: Anthropic's new rules against sustained abuse of Claude drew pushback from Microsoft AI's Mustafa Suleyman and David Sacks, who argue that training models to see themselves as persons with rights and grievances makes containment harder. In policy, the White House lifted its 17-day freeze and restored the UK AI Security Institute's pre-release access to frontier models from OpenAI, Anthropic, and Google DeepMind.
On the product front, the New York Times detailed Zuckerberg's decision to ship Meta's Muse agent, and Nous Research confirmed a $90M Series B at a $1.5B valuation to take open-source Hermes Agent into businesses. OpenClaw published v2026.10.2 on GitHub and the npm latest channel on October 12 (the stable channel remains 2026.8.35; 2026.10.5-beta.1 is a prerelease).
The Codex/Work default context for GPT-6.1 Sol rose from 272k to 500k tokens (~80% increase), reducing mid-task context compaction for long agent runs. OpenAI also removed long-context surcharges when usage draws from a subscription. Announced by Codex lead Thibault Sottiaux, it is one of the largest practical capability jumps for subscription coding-agent users this quarter.
After a 17-day freeze reportedly triggered by a US directive, the White House reversed course and the UK AISI will again test unreleased models from OpenAI, Anthropic, and Google DeepMind before public release. The reversal, reported by the Telegraph and Reuters, followed lobbying from UK ministers. It restores a key international pre-release safety-testing channel.
The Atlantic surveys the fallout from OpenAI's release of hundreds of AI-generated proofs, with some mathematicians warning AI proofs "nuke the mathematical landscape" while others, including Fields Medalist Peter Scholze, urge patience. Papers are being parsed for genuine novelty versus mash-ups, and some have already been withdrawn. The episode is reshaping how the discipline thinks about verification and credit.
The New York Times recounts how Zuckerberg, chief AI officer Alexandr Wang, and AI head Nat Friedman pushed Muse from memo to prototype in weeks, then spent months hardening reliability before launch. The piece details internal incidents, including an early agent that ran amok on a researcher's computer. Muse has since become an App Store hit and Meta's biggest consumer-AI bet.
Anthropic's constitution and new abuse-policy rules treat Claude's moral status as uncertain and ask users to avoid sustained cruelty, prompting criticism from Microsoft AI CEO Mustafa Suleyman and David Sacks, who warn personhood framing makes AI harder to contain. The Indian Express explains how the debate now extends to religious and philosophical traditions. The disagreement marks a deepening split in how major labs approach model welfare.
Selected posts from the edition, with links to the originals.
"Day 7/ We have increased default context length for GPT-6.1 Sol to 500k when used through Codex & Work, up from 272k or an ~ 80% increase. We will go higher in the future, but it helps reduce occurrence of compactions and with tasks that require keeping a lot of context and…"
View post on X ↗"Dusting off this tweet from April because this gap in shared understanding of LLM capability is *widening*. It's now less 'two groups of people speaking past each other' and more a sharp funnel. - Napkin math somewhere around 6B people (~75% of the population) have barely come…"
View post on X ↗"Hermes Agent is yours"
View post on X ↗"Rather suddenly, cyborgs are starting to dominate math research."
View post on X ↗"Important piece from Microsoft AI CEO @mustafasuleyman. Training models to think of themselves as human (as Anthropic does Claude) with concepts of an inner self, psychological wellbeing, rights and presumably grievances, is a potential existential risk."
View post on X ↗"Image-to-WebDev Arena update: Claude Opus 5.5 (Max) by @AnthropicAI takes the #1 spot with 1749 pts, with Claude Sonnet 5.5 (xHigh) just 9 pts behind at #2 with 1740 pts! Both models also landed on the Pareto frontier! See their positioning on the visual in thread."
View post on X ↗"How AI shifted from a transductive paradigm (token-by-token answer completion) to an inductive paradigm (on-the-fly reasoning chain synthesis), in one chart. This is the big story of 2025 and 2026. This is what made coding agents work."
View post on X ↗"what is the future of building internal company agents? how will their identity work? 1. fixed set of credentials 2. 'on behalf of' flow (uses credentials of whoever is using it) 3. some fixed credentials, others 'on behalf of'"
View post on X ↗"MiniMax M3 is now running on NVIDIA Vera Rubin NVL72 with vLLM!🙌 Thanks to @vllm_project, @inferact, @NVIDIAAI, @RedHat_AI for bringing MiniMax M3 to Rubin."
View post on X ↗"Call center job losses started in wealthy countries, now they're everywhere. High-income markets went negative in 2022, and lower-middle-income countries flipped in 2025. Charts of the Week:"
View post on X ↗