🔍 Read the full analysis: AI Lab Week: Anthropic, OpenAI, Google And Others In 10 Moves on ThorstenMeyerAI.com
Get hardware and tech essentials delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
A November 2–8 roundup from ThorstenMeyerAI.com counts 10 announcements from seven AI labs, covering agent memory, scientific work, coding, video and access to model weights. The source says it checked each item against lab posts, documentation or code, while noting that many performance figures are company-reported and three initial items did not pass its verification review.
The roundup’s leading theme is agents that retain information between sessions. It describes memory-related work from Anthropic, Meta and SpaceXSI, formerly xAI; two of the efforts include a review step for deciding what should be remembered. Anthropic’s November 6 engineering post, for example, describes memory as a combination of conversation records, model-written notes, linked information structures and a compact block loaded at the start of a session. Its authors also describe a process they call “dreaming,” which reviews older logs and updates memory.
The list also covers AI research and model access. Anthropic reported on Claude’s role in a biology project investigating vault particles, while Cognition introduced SWE-2 Flash but kept its weights private for now, citing potential cyber risk. Moonshot AI, by contrast, published the weights for Kimi K3.1 Pro on Hugging Face, under its own license. The source roundup stresses that published weights do not necessarily mean unrestricted use; the license and access terms matter.
Media tools and pricing are another part of the week’s activity. According to the source, Google’s video model can produce 15-second clips at $0.07 per second, while OpenAI’s image model can make animated GIFs. It also reports that ChatGPT and Codex will share one usage allowance on Plus, Pro and Business plans starting in December. These details are presented as developments reported in the roundup, not as independently audited performance or pricing comparisons.
Seven labs, ten moves Shaping how AI is used
A November roundup tracks changes in agent memory, research, model access, media tools and product usage. Its claims were checked against lab sources, though many performance figures remain company-reported.
Agents that remember
Anthropic, Meta and SpaceXSI featured memory work designed to carry information between sessions.
Conversation records and model-written notes help agents retain information.
Some approaches revisit older logs, update memory and review what should be remembered.
Access comes with conditions
Model availability depends on safety decisions and license terms.
- SWE-2 Flash weights remain private for now.
- A trusted-partner program is being developed.
- The company cited potential cyber risk.
- Kimi K3.1 Pro weights were published on Hugging Face.
- Use is governed by Moonshot’s own license.
Research, media and product use
Biology investigation
Anthropic described Claude’s role in research on vault particles; researchers had to catch mistakes, and the preprint was not yet public.
Per-second pricing
The roundup reports Google’s video model can make 15-second clips at $0.07 per second.
Animated GIFs
OpenAI’s image model can create animated GIFs, according to the roundup.
Shared allowance
ChatGPT and Codex are set to share one usage allowance on Plus, Pro and Business plans starting in December.
Read the evidence carefully
Source checks help establish what labs announced, but do not independently validate company results.
- Many performance figures are company-reported, not independent evaluations.
- Three initial items did not pass the roundup’s verification review.
- Licensing, infrastructure, safety and cost matter alongside benchmark scores.
Memory, Access and Usage Rules
Taken together, the announcements point to changes in how AI products are used and governed, rather than just new model releases. Persistent memory may let agents carry project information across sessions, but it also creates a need to review outdated or inaccurate notes. Anthropic’s own account of its memory system includes a process for revising stored information, while its research example describes mistakes that human researchers had to catch.
The contrasting decisions by Cognition and Moonshot AI show why model availability is not a simple open-or-closed choice. Cognition says it is limiting access to SWE-2 Flash weights because it believes the model’s capabilities could help attackers; Moonshot has released Kimi K3.1 Pro weights under a company-specific license. For developers and organizations, deployment choices depend on licensing, infrastructure, safety and cost—not only benchmark scores.
Usage limits and media-generation pricing also affect customers directly. A shared ChatGPT and Codex allowance could change how subscribers divide their use across products, while per-second video pricing gives users a stated cost basis. The roundup does not independently establish that these offerings deliver the reported performance or savings, and those distinctions matter when comparing products.
Products Worth Considering
Apple MacBook Air 13.6-inch Laptop with M5 Chip, 24GB Memory, 1TB SSD, Midnight
Apple MacBook Air 15-inch (2025) with M4 Chip, 24GB Memory, 512GB SSD, Midnight
Samsung Galaxy Tab A11+ 11-Inch Wi-Fi Tablet, 6GB RAM, 128GB Storage, Gray
As an affiliate, we earn on qualifying purchases.
How the Week Was Verified
The source article, published by Thorsten Meyer on November 8, frames its list as a review of seven days of AI-lab announcements. It says every item was checked against an original lab post, documentation or code. That approach helps distinguish published product details from unsupported online claims, but checking a company’s source material does not independently validate the company’s test results.
The article specifically warns that results such as memory scores, coding benchmarks and claimed time savings are self-reported. It says three items on its initial list did not survive its source check and were excluded or explained at the end. The material provided here does not identify those three items or include the full details of all 10 moves, so this account focuses on the announcements and figures present in the supplied source.
For Anthropic’s biology report, the source says the project took under four months and involved searches across about 15 billion predicted proteins. Researchers tested 44 candidates, with seven of 15 proteins reported as soluble forming vault-like particles. Anthropic also acknowledged a protocol error that cost two weeks and an image map that blended with its mirror image. The source says the related preprint had been submitted but was not yet public at the time of the roundup.
Limits of the Reported Evidence
The supplied source does not provide the names and full descriptions of all 10 announcements, nor the identities of the three entries that failed its verification check. It also does not include independent tests of the labs’ benchmark results, pricing claims or performance comparisons. Those figures should be read as reported company measurements unless corroborated by separate evaluations.
Some details remain time-sensitive. The source says the Anthropic biology preprint was not yet public, and Cognition’s planned trusted-partner program for SWE-2 Flash weights was still being built. It does not specify the final eligibility rules or timing for that program. The provided text also says ChatGPT and Codex sharing one allowance begins in December but gives no exact date or detailed allowance terms.
Upcoming Access and Usage Changes
The next stated milestone is the planned December change to ChatGPT and Codex allowances for Plus, Pro and Business subscribers. The source does not give the precise launch date or explain how the shared allowance will be calculated, so users will need the product’s published plan details for those terms.
Cognition says it is developing a process to share SWE-2 Flash weights with trusted partners, but the source gives no launch schedule. Moonshot AI’s weights are available under its own license, and the Anthropic biology research may become easier to assess once its submitted preprint is public. Further independent testing will be needed to compare the reported scores and operating costs across labs.
Key Questions
What is the main development in the roundup?
ThorstenMeyerAI.com summarized 10 announcements from seven AI labs during November 2–8, 2026, spanning agent memory, scientific research, coding, video generation, model weights and product usage limits.
Were all the reported performance figures independently verified?
No. The source says it checked announcements against lab posts, documentation or code, but identifies many scores and measurements as company-reported. It does not provide independent tests of those results.
What is the difference between Cognition’s and Moonshot AI’s model releases?
Cognition introduced SWE-2 Flash but said its weights would not be publicly released yet, citing cyber risk and a planned trusted-partner program. Moonshot AI published Kimi K3.1 Pro weights under Moonshot’s own license.
When will ChatGPT and Codex share an allowance?
The roundup says the shared allowance for Plus, Pro and Business plans is due to begin in December. It does not state an exact date or provide the detailed usage rules.
Source: ThorstenMeyerAI.com
