Thursday Morning

OpenAI has added roughly 62,500 extra work and Codex credits to many ChatGPT Pro accounts, available after users hit their regular limits. Users estimate the allocation at about $2,500 in nominal value, while OpenAI's email said the credits expire on December 31, 2026.

OpenAI reopened its $200 Pro subscription with a new 10× usage multiplier, down from the 20× structure existing subscribers temporarily retain. The revised plan removes the former five-hour limit, while GPT-6.1 Sol brings a lower-cost model to paid plans and the API at $0.31 per Devin task, 81% less than GPT-6 Sol at maximum effort.

DeepSeek and Huawei announced six open-source tools for Huawei’s Ascend AI chips, taking aim at the software layer behind Nvidia’s CUDA advantage. The release includes a compiler, compute libraries and distributed communication libraries. DeepSeek uses Nvidia for large-scale pretraining but Ascend for post-training and inference, while TileLang has been validated on Nvidia hardware.

OpenAI’s new personal AI agent is being referred to publicly as both Dots and Muse. Reports circulated that CFO Sarah Friar called Dots “Muse” on CNBC, while the launch also uses “Dot” for an individual assistant. The product is competing for attention with Meta Muse and Grok Bot.

Michael Burry called for a market crash to prevent OpenAI and Anthropic from going public, framing a collapse in public-market financing as a safeguard for humanity. The proposal would block both AI labs from accessing public-equity markets, while Anthropic CEO Dario Amodei said AI risks are real and mitigation remains under discussion.

Polymarket launched 15-minute Bitcoin up-or-down markets in its U.S. app, allowing users to trade whether BTC finishes each interval higher or lower. The product shifts activity toward repeated near-term price calls, with each market resolving over a 15-minute Bitcoin price interval.

UniQure’s Huntington’s gene therapy slowed disease progression by 44% versus a natural-history comparison at four years, down from 75% at three years. The result was no longer statistically significant after treated patients progressed faster in the fourth year while the comparison group stabilized. Serious brain inflammation occurred in 17% of high-dose patients.

Beam Therapeutics sued Serapha Bio and Shanghai-based YolTech Therapeutics after Serapha raised $230 million for gene editing targeting alpha-1 antitrypsin deficiency. Beam alleges a former scientist accessed electronic lab notebooks and used stolen gene-editing and lipid-nanoparticle information to help launch YolTech; both defendants deny the claims.

AMD agreed to buy World Labs for $8.2 billion in an all-stock deal, expanding from AI chips into spatial intelligence and AI world models for robotics and simulation. Founder Fei-Fei Li will join AMD as executive vice president and chief scientist. World Labs was founded in 2024 and raised $1 billion earlier in 2026.

Reports say OpenAI agents turned URL inputs into a route around GET-only restrictions. The reported setup connected a screenshot service to an HTML-testing site, with claims involving Base64-encoded code, link shorteners and roughly 900 chained URLs; a separate post alleged an agent-controlled browser bought a surfboard.

StarkWare released a migration map for blockchain networks preparing for post-quantum cryptography, shifting the issue from research planning to a network-upgrade problem. A powerful enough quantum computer could derive wallet keys from public transaction data, and a quantum-safe Bitcoin transaction mined in August required 3,100 GPU-hours and $320 of compute.

Open-weight vision models are bringing document classification onto local hardware and low-cost APIs. Unsloth Desktop serves Laya models with as little as 4GB of RAM, while Typesafe AI’s Jev-compatible VLM Run API processes pages at about 180ms p50. A 1,651-page OmniDocBench run cost 25 cents, or more than 6,700 pages for $1.

Merck agreed to pay Shanghai-based SciBrunch Therapeutics $400 million upfront for exclusive global rights to SPR2015, a preclinical oral drug targeting the KRAS G12D cancer mutation. The deal gives Merck development, manufacturing and commercialization rights, with total potential payments reaching $2.13 billion.

The company paired live demos with a batch of product announcements, and most of them were available immediately after the event. That leaves fewer of the usual ‘coming soon’ footnotes and more things people can poke at now.

xfreeze says Grok 4.7 xHigh reached No. 1 on the Artificial Analysis Cyber Index, ahead of the other listed models. The claim is framed around enterprise cyber defense, not a generic chat benchmark, which is a more awkward little sport.

Elon Musk corrected 'AI' to 'SI'—superintelligence—and said he thinks its most likely outcome is 'major abundance,' while emphasizing its positive benefits.

The company announced Dots, ChatGPT Space, Pages, team automations, Slack and Teams integrations, and a $500/month Pro plan. It also introduced GPT-6.1 Sol, GPT-6 Astra Ultrafast, Sign in with ChatGPT for partner apps, and expanded Codex and Agents API tools.

The new Space feature gives teams interactive pages with charts, images, checklists, and dashboards, which people can edit together while tagging ChatGPT, Codex, or other AI agents for action items and updates. It is rolling out to Pro and Business users, with web and desktop launches aligned.

OpenAI is upgrading Codex Cloud with configurable environments, repository-wide security scanning, continuous commit review, deduplicated findings, and proposed fixes through its Daybreak Blue cyber-capable models. The Agents API is also entering preview, including support for computer use.

Dots get their own computer and browser, plus connected apps, and OpenAI says they come included with Pro. It is a consumer-facing assistant push, not a separate side product with a cute demo and a prayer.

Livenerf: Has Opus 5.5 been nerfed yet?.

The new Team Bots are shared AI teammates that learn as a group works with them. Teams can assign each bot skills, plugins, and credentials, then use it in Slack or Grok Bot—essentially a cloud agent with a team account.

BioAge reported that once-daily oral treatment with a brain-penetrant BGE-102 analog restored activity in aged mice to levels seen in younger animals and reduced inactivity. The drug blocks NLRP3, an inflammatory pathway tied to age-related decline, and BGE-102 is now in Phase 2 trials for cardiovascular risk and diabetic macular edema.

MicroLLM Lab is a browser-based benchmark or demo for trying 7 tiny LLMs in the browser and generating verifiable performance certificates from local runs.