🔍 Read the full analysis: What Seven AI Labs Did This Week: 10 Moves In Review on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
A November 8 roundup from ThorstenMeyerAI.com counted 10 announcements by seven AI labs from November 2–8, covering agent memory, scientific research, model access, video and subscription limits. The source says it verified the items against lab posts, documentation or code, while warning that many performance and cost figures are company-reported and that three items on its initial list did not pass verification.
Seven AI labs made 10 announcements during the week of November 2–8, according to a review published November 8 by ThorstenMeyerAI.com. The roundup, which says it checked each item against a lab post, documentation or code, highlights work on agent memory, scientific research, model access and generative media, alongside a forthcoming change to ChatGPT and Codex subscription limits.
The roundup identifies memory for AI agents as a prominent theme. Anthropic published an engineering account of Claude’s memory system, describing session transcripts, model-written notes, larger linked structures such as wikis, and a small amount of context loaded at the start of a session. The account also describes a process called “dreaming” that reviews previous logs and updates stored information. Anthropic reported that memory improved Opus 5.5’s mean reward from 0.655 to 0.838 on one database task; those figures come from the company’s own test, not an independent evaluation.
Anthropic also described using Claude in work on the structure of a bacterial vault particle. The company said its team searched roughly 15 billion predicted proteins, tested 44 candidates and found that seven of 15 proteins that came out soluble formed vault-like particles. The project took less than four months, according to the post. Anthropic also disclosed setbacks, including a protocol error that cost two weeks and a map that combined a structure with its mirror image. The source says the related preprint had been submitted but was not yet public.
Other entries covered competing choices about model access. Cognition released SWE-2 Flash for use in its Devin coding agent but kept its weights private, citing cybersecurity concerns. Moonshot AI, by contrast, published the weights for Kimi K3.1 Pro on Hugging Face under its own license. The roundup also points to Google’s video model, which it says produces 15-second clips at $0.07 per second, and OpenAI’s image model, which it says can generate animated GIFs. It reports that ChatGPT and Codex will share one usage allowance on Plus, Pro and Business plans beginning in December.
Seven Labs, Ten Moves What changed this week
A November 8 roundup reviewed AI announcements from November 2–8, spanning agent memory, research, model access, media generation and subscription limits.
Memory and research
Anthropic described persistent agent memory and a biology research effort, including reported setbacks.
Claude can carry context forward
Stored notes and linked information can help an agent retain useful details across sessions; those notes still need review.
AI-assisted biology work
Anthropic reported searching predicted proteins and testing candidates. Its vault-structure preprint had been submitted but was not yet public.
Model access choices
The releases show that access depends on each lab’s distribution terms.
- Cognition released SWE-2 Flash for its Devin coding agent.
- It kept the weights private, citing cybersecurity concerns.
- Moonshot AI published Kimi K3.1 Pro weights on Hugging Face.
- The weights are under Moonshot’s own license.
Products and subscriptions
The roundup also covered new media capabilities and a planned change for subscribers.
Google’s video model was reported to make 15-second clips at $0.07 per second.
OpenAI’s image model was reported to generate animated GIFs.
ChatGPT and Codex are reported to share one allowance on Plus, Pro and Business plans starting in December; the amounts were not provided.
How to read the roundup
The publication says it checked included items against lab posts, documentation or code.
- Many performance and cost figures are company-reported, not independently verified.
- The supplied material does not list all seven labs, all ten moves or the three removed items.
- Review licenses and plan terms before making deployment or usage decisions.
Memory, Access and Usage Limits
Taken together, the announcements point to practical choices facing AI users and businesses: whether a system can retain useful information, whether its model weights can be run independently, and how access is priced or capped. Memory can make an agent more useful across sessions, but it also creates a need to check whether stored notes are accurate and appropriate. Anthropic’s own account of errors in its biology work underscores that human review remains part of the process.
The contrasting releases from Cognition and Moonshot show that “open” can refer to different things. A model may be built using an open-weight component while its final weights remain restricted; a published weight file, meanwhile, may still come with a custom license. Organizations considering deployment need to review the actual access terms, not infer them from a model’s name or technical lineage.
For subscribers, the reported December change could affect how people allocate use between ChatGPT and Codex. The source does not provide the new allowance amounts, so the practical effect cannot yet be determined. And while reported prices, benchmark scores and time savings offer clues about product direction, they are not independently verified results in the material supplied.
Popular right now
Breville Barista Express Espresso Machine with Built-In Grinder, Brushed Stainless Steel
De'Longhi COM630BK All-in-One Coffee and Espresso Machine
KitchenAid Classic Series 4.5-Quart Tilt-Head Stand Mixer, Onyx Black
As an affiliate, we earn on qualifying purchases.
How the Weekly Review Was Checked
The source presents this as a weekly roundup rather than a single product launch. It says each of the 10 included items was checked against a primary source, such as a company announcement, product documentation or code. It also says three items from its first list were removed because they did not survive that check, though the supplied material does not identify those items or explain each removal individually.
The review covers announcements from November 2 through November 8, 2026. It says Anthropic and OpenAI were among the companies covered and that neither reviewed nor sponsored the roundup. The source also warns that figures such as model scores, cost comparisons and claimed time savings are often the labs’ own measurements. They should be read as company-reported results, not as independently established comparisons.
“Memory was the theme.”
— ThorstenMeyerAI.com’s roundup
Limits of the Reported Evidence
The supplied material names only some of the 10 announcements and does not provide a complete list of all seven labs or all 10 moves. It also does not identify the three items removed from the original list. The roundup says it checked its included items against primary sources, but the source text provided here does not include links or supporting documentation for every individual claim.
Many numerical results are attributed to the companies themselves. The available material does not establish whether Anthropic’s task scores, Cognition’s coding and security benchmarks, or the stated video price were independently replicated. It also does not specify the usage quantities for the planned shared ChatGPT and Codex allowance, or how the change will affect individual customers. The Kimi weights are described as released under Moonshot’s own license, but the supplied text does not reproduce the license terms.
December Plan Changes and Releases
The clearest dated next step in the roundup is the planned December start for a shared ChatGPT and Codex allowance on Plus, Pro and Business plans. OpenAI’s exact limits and any account-specific details are not included in the supplied material. Cognition, meanwhile, says it is building a program to share SWE-2 Flash weights with trusted partners, but gives no launch date.
Further scrutiny will depend on the publication of supporting research and fuller product documentation. The source says Anthropic’s vault-structure preprint had been submitted but was not yet public; its availability could allow outside researchers to examine the work. Users and organizations will also need to consult the relevant license and plan terms as the releases and usage changes take effect.
Key Questions
What is the main development covered in this report?
ThorstenMeyerAI.com published a review of 10 AI-lab announcements from November 2–8, 2026, grouping them around agent memory, scientific research, model access, media generation and subscription usage.
Which companies’ announcements are described in the supplied material?
The excerpt names Anthropic, Cognition, Moonshot AI, Google and OpenAI, and refers to SpaceXSI (formerly xAI) in its summary of memory-related work. It does not provide the full list of seven labs or all 10 announcements.
Are the performance and cost figures independently verified?
Not on the basis of the supplied material. The roundup says many figures, including benchmark scores and product measurements, are reported by the companies themselves and should be treated as signals rather than independent proof.
What is changing for ChatGPT and Codex subscribers?
The source says ChatGPT and Codex will share one usage allowance on Plus, Pro and Business plans starting in December. It does not state the allowance amounts or explain how the change will affect each plan.
Why are Cognition’s model weights not public?
Cognition says it is withholding SWE-2 Flash’s weights because its coding and security capabilities could be misused by attackers. The company says it is developing a trusted-partner access program, but no timing is given in the supplied material.
Source: ThorstenMeyerAI.com
