AI news6
METR finds Anthropic’s benchmark progress running about 1.9 times faster
METR’s 5 November assessment estimates that Anthropic’s AECI benchmark scores have improved at 1.9 times their pre-October 2025 rate, with a 95% confidence interval of 1.5–2.8 times. The team judges AI a major contributor, while acknowledging uncertainty and alternative explanations. Anthropic redacted much of the evidence underlying the assessment.
metr.org ↗Why it matters
This new independent assessment goes beyond earlier Anthropic product announcements, while showing how limited public evidence complicates claims of AI accelerating AI research.
Thinking Machines releases Inkling 2 Small through Tinker
Thinking Machines introduced Inkling 2 Small on 5 November, with 594 billion total parameters and 17 billion active per token. The model supports image inputs and is available for inference and fine-tuning through Tinker. Its weights remain closed, with an Apache 2.0 release planned for next week after external safety testing.
thinkingmachines.ai ↗Why it matters
This second model generation extends the earlier Inkling release, but self-hosting still depends on the promised weights becoming available.
Google cuts Omni 1.2 Flash video pricing to US$0.07 per second
Google’s 5 November Gemini Omni 1.2 Flash release generates and edits clips up to 15 seconds long, increased from 10 seconds. The Gemini API price for 720p video is US$0.07 per second, a 30% reduction from Omni 1.1. Google says the model is generally available through the Gemini API and Google Cloud, with multi-turn editing and stronger reference adherence.
blog.google ↗Why it matters
The new video API release adds longer clips at lower cost beyond Google’s previously covered models, reducing the expense of testing creative variants.
OpenAI Images 2.6 adds GIFs and reports 50% faster generation
OpenAI’s 4 November Images 2.6 release adds GIF creation in ChatGPT, Codex and the API, alongside generation it says is 50% faster. Developers can choose Flare for speed or Sunburst for higher quality, with image markup and SVG exports also reaching the API. The company says low-quality Flare images can take as little as six seconds.
openai.com ↗Why it matters
This new visual-production release adds faster iteration and animation beyond OpenAI’s previously covered coding and cyber tools.
Perplexity Teammates takes on ongoing work for Max subscribers
Perplexity’s Teammates agents can take standing responsibilities, coordinate other agents and work in the background around the clock. Each keeps a persistent thread and alerts the user when a decision is needed. The current product page lists availability for Max subscribers on the web, with iOS and Android coming later.
engadget.com ↗Why it matters
This ongoing-work layer extends the earlier Perplexity Computer offering by keeping responsibilities active between individual requests.
Aina opens US$399 preorders for an AI ring shipping in early 2027
Aina’s 5 November launch introduces a ring with an OLED display and thumb gestures for approving AI actions, taking notes and controlling calls or Mac apps. The current site lists a starting preorder price of US$399 and shipping in early 2027. It pairs with iPhone and Mac and claims up to 18 hours of battery life.
aina.com ↗Why it matters
The new ring puts a visible approval step on the body, offering a distinct interface for acting on AI suggestions.
AI in the nonprofit sector2
AACN awards two US$25,000 grants for AI in nursing education
AACN’s Foundation for Academic Nursing announced two US$25,000 Faculty Scholars grants on 6 November, selected from more than 100 applications. Tonychris Nnaka at UNT Health will study AI-powered holographic patient simulations, while Columbia’s Natalya Pasklinsky will study AI-enabled preparation for communication and documentation training. The next funding call is scheduled for May 2027.
newswise.com ↗Why it matters
The awards fund two specific teaching projects whose results can help nursing schools assess where AI simulation improves training.
LDRI’s climate AI intake closes temporarily on 15 November
The Local Development Research Institute’s RICH Hub is seeking African-led or Africa-based climate innovations that use AI or data and have reached pilot stage or beyond. Selected projects may receive technical support, partnerships and potential scaling grants up to US$200,000, subject to assessment and resources. The live expression-of-interest form closes temporarily on 15 November for first-cohort review, with no cutoff time or timezone specified.
climatehub.developlocal.org ↗Why it matters
This live climate-innovation intake is separate from the earlier AI4D compute programme, and offers support without guaranteeing a grant.