🔍 Read the full analysis: From OpenAI To Moonshot: 10 Moves In AI Lab Week on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
A Nov. 8 roundup from ThorstenMeyerAI.com describes ten announcements from AI labs during the week of Nov. 2–8, including agent memory work, a biology project, model releases and changes to product usage limits. The article says it checked items against lab posts, documentation or code, while noting that many performance figures are company-reported and three items on its initial list did not pass its source checks.
ThorstenMeyerAI.com published a roundup on Nov. 8 of ten announcements from AI labs during the week of Nov. 2–8, covering agent memory, scientific research, model access, media generation and subscription limits. The report says it checked each item against a lab post, documentation or code, and warns that many reported scores and cost figures come from the companies themselves.
The roundup’s main theme is agents that retain information across sessions. It describes an Anthropic engineering post about four layers of Claude’s memory: session transcripts, model-written notes, linked structures such as wikis, and a small context block loaded at the start of a session. Anthropic also described a review process called “dreaming,” which checks older logs and updates memory. The source says Anthropic reported a rise in mean reward from 0.655 to 0.838 on one database task when memory was used; that result is from the company’s own test.
Other entries span science and model availability. Anthropic described using Claude in research on the structure of a bacterial vault particle, while acknowledging errors that researchers had to catch; the source says the related preprint had been submitted but was not yet public. Cognition released SWE-2 Flash but said its weights would not initially be public, citing cyber risk. In contrast, Moonshot AI published the weights for Kimi K3.1 Pro on Hugging Face. The roundup lists the model at 2.8 trillion total parameters and says it is distributed under Moonshot’s own license, rather than a standard open-source license.
The report also points to changes in products and usage. It says Google’s video model generates 15-second clips at a stated price of $0.07 per second, and that OpenAI’s image model can make animated GIFs. It reports that ChatGPT and Codex will share one usage allowance on Plus, Pro and Business plans starting in December. These are details as presented in the roundup; independent verification of company performance and pricing figures is not described.
From OpenAI to Moonshot 10 moves in AI lab week
A Nov. 8 roundup covers AI lab announcements from Nov. 2–8, spanning agent memory, science, model access and product limits.
Memory that carries forward
Anthropic describes a layered system for retaining and reviewing information across sessions.
Transcripts preserve what happened in earlier sessions.
Model-written notes and linked structures such as wikis organize context.
A process called “dreaming” checks and updates stored memory.
Model access comes with terms
The roundup contrasts restricted weights with a published release under a company-specific license.
- Weights not initially public
- Cyber risk cited
- Trusted-partner access proposed
- Weights published on Hugging Face
- Moonshot’s own license
- Not a standard open-source license
Science still needs human review
Anthropic described Claude’s role in research on a bacterial vault particle, while noting errors researchers caught.
Biology project
The related preprint was submitted but not yet public in the roundup.
Errors were caught
Researchers identified a protocol error and a map blending a structure with its mirror image.
Products and limits shift
The roundup reports changes that could affect media workflows, costs and how customers use subscriptions.
| Product | Reported change | What to check |
|---|---|---|
| Google video model | 15-second clips at $0.07 per second | Pricing and real-world costs |
| OpenAI image model | Can make animated GIFs | Fit for the intended workflow |
| ChatGPT and Codex | Shared allowance planned from December | Exact limits and eligible users |
Memory and Access Shape AI Use
The announcements matter because they touch practical choices for companies and users: whether AI systems can retain useful project information, how much human review scientific work still needs, and whether customers can run a model themselves or must use a hosted service. Memory can preserve context, but it also creates a need to check whether stored notes are accurate and appropriate to reuse.
The differing release choices also show that “open” does not describe every aspect of a model. Cognition’s reported decision to restrict SWE-2 Flash weights contrasts with Moonshot’s publication of Kimi K3.1 Pro weights under a company-specific license. Access, licensing and safety conditions can affect whether organizations can inspect, adapt or host a model, and those terms should be checked directly rather than inferred from a model’s lineage.
For businesses, the reported move toward longer video generation and shared product allowances may affect costs and workflows. But lab-reported benchmarks, prices and savings do not by themselves establish how a system will perform in another organization’s setting. The roundup presents these figures as signals, not independent proof.
Products Worth Considering
Bose QuietComfort True Wireless Earbuds with Active Noise Cancellation, Bluetooth 5.3, USB-C Charging, IPX4, Black
Samsung Galaxy Tab A11 8.7-Inch Wi-Fi Tablet, 64GB, 4GB RAM, 90Hz, Gray, with 20W Charger
Jackery Solar Generator 1000 v2 with Two 100W Solar Panels, 1024Wh LiFePO4 Battery
As an affiliate, we earn on qualifying purchases.
A Week of Lab Announcements
The source frames the ten announcements as a seven-day snapshot rather than a single coordinated industry event. Its stated coverage period is Nov. 2–8, 2026, and it says the entries were checked against primary material such as company posts, documentation or code. Anthropic and OpenAI are among the companies covered; the publisher says neither reviewed nor sponsored the roundup.
The report also describes a separate Anthropic analysis of 207,049 scientific sessions from August and September. According to the source, data analysis accounted for 33.7% of Claude sessions, writing for 32%, and computer science for 32.7% of Claude Code sessions. These figures describe Anthropic’s own product data and are not presented as a measure of all scientific or coding work.
The source says three items on its original list did not survive its verification check, but the material supplied here does not identify those items or explain the checks in detail. That qualification limits what can be concluded about the full set of developments in the week.
“The authors call one review step “dreaming.””
— ThorstenMeyerAI.com, describing Anthropic’s memory work
Company Metrics Need Outside Checks
The supplied material does not establish independent replication of the companies’ benchmark results, pricing comparisons or performance claims. Anthropic’s memory score, Cognition’s coding and security scores, and the video pricing are presented as company-reported figures. They should not be treated as directly comparable without shared testing methods and fuller measurement details.
The biology project’s preprint was described as submitted but not yet public, so its methods and findings could not be evaluated from that paper at the time covered. The source also notes that researchers caught a protocol error and a map that blended a structure with its mirror image. It does not provide the names of the three rejected items or enough detail to assess why they failed verification.
Some forward-looking details are also incomplete. The roundup says ChatGPT and Codex allowances will be shared from December, but the provided material does not state the exact allowance, how current limits will change, or whether all eligible users will be affected in the same way.
Watch for Terms and Follow-Up Data
The next developments to watch are the publication of the biology preprint, further information about Cognition’s proposed trusted-partner access to SWE-2 Flash, and the practical details of the shared ChatGPT and Codex allowance planned for December. Those details will help clarify what users can access and how the changes affect costs.
For the reported model and product claims, independent evaluations and user experience will provide a stronger basis for comparison than company figures alone. The roundup also says three items were removed after source checks, but does not name them in the supplied material; any fuller account would need to identify those entries and explain the verification outcome.
Key Questions
What is the actual news in the roundup?
It is a weekly recap, published Nov. 8, 2026, describing ten AI lab announcements from Nov. 2–8 across memory, science, model access, media generation and product usage.
Did Anthropic independently prove its memory results?
No independent replication is cited in the supplied material. The reported rise in mean reward—from 0.655 to 0.838 on one database task—is attributed to Anthropic’s own test.
Are Kimi K3.1 Pro’s weights open source?
The roundup says Moonshot AI published the weights on Hugging Face, but under Moonshot’s own license, not a standard open-source license. Users should review the license terms for their intended use.
Why did Cognition keep SWE-2 Flash’s weights private?
According to the roundup, Cognition cited cyber risk, saying the model’s capabilities could also assist attackers. It said it was working on a program to share the weights with trusted partners.
What remains unverified or unknown?
The supplied material does not provide independent checks of the labs’ benchmark and pricing claims, identify the three items removed from the initial list, or give full details of the shared ChatGPT and Codex usage allowance planned for December.
Source: ThorstenMeyerAI.com
