AI News Nov 15, 2026
Language: DE / EN / ZH / FR / ES / PT / JA / KO
AI-generated: summaries written by AI from the linked sources. How we use AI
Key Takeaways
- Meituan released the weights for LongCat-2.5, a 1.6-trillion-parameter model built for AI agents and trained entirely on Chinese hardware with no mention of Nvidia GPUs. It scores …
- Anthropic's Claude Fable 5.2 generates images by writing Python code that places hundreds of thousands of simulated brushstrokes on a virtual canvas rather than computing pixels, c…
- OpenAI published a candidate nonperturbative lattice definition of the Standard Model and says GPT-6 Astra helped turn the idea into a detailed proof. The work was shared on OpenAI…
- Nidus: $30M (Pre-Seed and Seed)
- Point Datasette's LLM plugin at a local OpenAI-compatible server (e.g. an api_base like http://localhost:1234/v1 with a model name) to run a Datasette Agent ins…
Why Today Matters
- China AI computeMeituan's LongCat-2.5, a 1.6-trillion-parameter model the company says was trained entirely on Chinese hardware, lands near Claude Opus 5 on Terminal Bench, while an Epoch AI analysis puts Huawei's Horinger data center at roughly 7% of SpaceX's Colossus 2 compute.
- Model capabilitiesAnthropic's Claude Fable 5.2 renders images through simulated brushstrokes rather than pixels, an approach one employee says took about 700,000 strokes for a single scene, while OpenAI says GPT-6 Astra helped develop a lattice definition of the Standard Model.
- Export controlsCongress is weighing tighter AI chip export limits after a scandal involving former Nvidia CEO Jensen Huang, and the US plus Pax Silica allies including Japan, the UK and India agreed to a joint list of technologies to protect from China.
- AI labor and limitsAustralia's central bank sees early signs that AI is hitting 15- to 24-year-olds in exposed roles like software development, though author Mike Major cautions other factors could explain the drop, while UniPat AI's SPU benchmark found no agent configuration above 0.5.
- Product and fundingApple reportedly plans to unveil smart glasses at WWDC and ship by end of 2027, centered on Siri AI integrations with third-party iPhone apps, as funding flows to applied AI: Nebuly raised a $21M Series A and Nidus $30M for defense manufacturing robotics.
What are the top AI breakthroughs?
This Nov 15, 2026 covers 10 curated AI news items spanning technology, research, and product developments. Meituan released the weights for LongCat-2.5, a 1.6-trillion-parameter model built for AI agents and trained entirely on...
Meituan released the weights for LongCat-2.5, a 1.6-trillion-parameter model built for AI agents and…
Meituan released the weights for LongCat-2.5, a 1.6-trillion-parameter model built for AI agents and trained entirely on Chinese hardware with no mention of Nvidia GPUs. It scores 88.4 points on Terminal Bench, landing just behind Claude Opus 5, and the sample deployment runs on 194 Huawei Ascend chips.
Anthropic's Claude Fable 5.2 generates images by writing Python code that places hundreds of thousan…
Anthropic's Claude Fable 5.2 generates images by writing Python code that places hundreds of thousands of simulated brushstrokes on a virtual canvas rather than computing pixels, complete with bristle texture and paint application. According to Anthropic employee Jake Eaton, a single photorealistic parking garage scene took roughly 700,000 strokes.
OpenAI published a candidate nonperturbative lattice definition of the Standard Model and says GPT-6…
OpenAI published a candidate nonperturbative lattice definition of the Standard Model and says GPT-6 Astra helped turn the idea into a detailed proof. The work was shared on OpenAI's new science blog, which publishes findings, methods, and firsthand accounts from researchers using AI in scientific discovery.
Following a scandal involving now-former Nvidia CEO Jensen Huang and a suspected Chinese spy, Congre…
Following a scandal involving now-former Nvidia CEO Jensen Huang and a suspected Chinese spy, Congress is looking to further restrict AI chip exports, with both Democrats and Republicans backing various measures and bills. Lawmakers have also demanded that Huang and other Nvidia leaders testify before Congress.
According to Bloomberg's Mark Gurman, Apple plans to unveil smart glasses at WWDC and ship them by t…
According to Bloomberg's Mark Gurman, Apple plans to unveil smart glasses at WWDC and ship them by the end of 2027, focusing heavily on integrating third-party iPhone apps with Siri AI. Apple's just-announced Siri AI integrations are described as the centerpiece of the effort.
Australia's central bank sees early signs that AI is hitting young entry-level workers in highly exp…
Australia's central bank sees early signs that AI is hitting young entry-level workers in highly exposed jobs like software development and customer service. Since ChatGPT launched, job postings and employment among 15- to 24-year-olds in those fields have dropped more sharply than among older workers, though author Mike Major cautions that business cycle shifts and remote work trends could also explain the numbers.
An Epoch AI analysis finds Huawei's Horinger data center, among China's largest for AI, has roughly …
An Epoch AI analysis finds Huawei's Horinger data center, among China's largest for AI, has roughly 7% of the compute of SpaceX's Colossus 2 and may stay far behind until at least 2028. Huawei, China's biggest AI chip developer, is drawing on China's abundant energy to build data centers.
The US and Pax Silica allies including Japan, the UK, and India agreed to create a joint list of AI …
The US and Pax Silica allies including Japan, the UK, and India agreed to create a joint list of AI technologies and supply chains to protect from China, according to the Financial Times. The pact comes ahead of Donald Trump's meeting with Xi Jinping at the Asia-Pacific Economic Cooperation forum in Shenzhen.
UniPat AI's SPU benchmark asks AI agents to rank the outcomes of 4,207 real wet-lab experiments, and…
UniPat AI's SPU benchmark asks AI agents to rank the outcomes of 4,207 real wet-lab experiments, and none of the 13 configurations tested scored above 0.5. The best configuration averaged 0.486 out of 1, highlighting how far agents remain from reliably predicting experimental results.
HyperS2C, formed from S2C and HyperSilicon, launched a '5+1' digital front-end EDA toolchain along w…
HyperS2C, formed from S2C and HyperSilicon, launched a '5+1' digital front-end EDA toolchain along with the HyperHybrid platform and two AI agents for verification. The launch adds AI-driven verification agents to a full front-end chip design toolchain.
Top AI Videos This Week
2 curated YouTube videos about AI developments.
Based on a transcript excerpt: This AI news roundup covers Google's release of Gemini 4 Carbon and t…

Based on a transcript excerpt: This AI news roundup covers Google's release of Gemini 4 Carbon and the more efficient Gemini 4 Hydrogen, plus new open-source text-to-speech models with real-time, interruptible voices. It also highlights ByteDance's Seedream 5.5 image model, a new protein/DNA predictor that reportedly beats AlphaFold 3, DeepMind's Alpha Genome 2, and various open 3D generators and video tools.
Based on a transcript excerpt: The top scorer of South Korea's certified tax accountant exam, surnam…

Based on a transcript excerpt: The top scorer of South Korea's certified tax accountant exam, surnamed Han, publicly admitted to cheating using generative AI and turned himself in to police. He confessed to filming exam questions with a small camera and receiving AI-generated answers during both rounds of the tax accountant exam, and also during the first round of the CPA exam; his exceptionally high scores had raised suspicions.
What are the latest AI investment signals?
Latest AI investment signals: 2 funding rounds, 2 market updates, and 0 M&A transactions.
Primary Market – Funding Rounds
Secondary Market – Market Updates
M&A – Mergers & Acquisitions
No M&A data.
What are practical AI tips this week?
4 practical AI tips curated from Reddit communities and expert blogs. datasette-llm 0.1a11 adds support for OpenAI-compatible endpoints, so Datasette can be configured to...
Productivity
datasette-llm 0.1a11 adds support for OpenAI-compatible endpoints, so Datasette can be configured to use an LLM hosted anywhere that speaks the OpenAI pseudo-standard web protocol, including local models run with LM Studio, Ollama or llama.cpp.
Point Datasette's LLM plugin at a local OpenAI-compatible server (e.g. an api_base like http://localhost:1234/v1 with a model name) to run a Datasette Agent instance powered by your own local model.
Productivity
datasette-apps 0.4a0 adds support for linking: apps can read #fragments from the URL, and supported URL schemes are reported back to Datasette Agent so it can help write apps that link to each other.
Split functionality across small linked apps — e.g. one app that shows details about a specific record and a separate app that searches records and links to their representations in the first app — using URL #fragments to pass state between them.
Prompt Tips
datasette-agent 0.6a0 adds an agent prompt directly to the Datasette homepage, with support for hash-mentions of tables and queries.
When asking an agent questions about your data, use hash-mentions of tables and queries in the prompt to point the agent at the specific data you mean.
Productivity
The newsletter argues that eval fluency is changing what businesses will pay for AI, and suggests using it in your next purchase while working out how much of the benefit your business gets to keep.
Before an AI purchase, apply eval fluency — assess and compare measured AI performance — and work out how much of the resulting benefit your business retains versus passes through in price.