AI News Nov 14, 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
- Nvidia CEO Jensen Huang has exited the company, prompting Asian governments to react: some moved to confirm chip supply commitments, while others are monitoring impacts on local pr…
- Google's Gemini 4 Carbon and Anthropic's Fable 5.2 have displaced OpenAI's GPT-6.1 Sol from the top of the MathArena math benchmark, the first time in a while no OpenAI model leads…
- Runway released Praxis-1, a robotics model with open weights and source code under the Apache 2.0 license. The model transfers knowledge from video training to robot control and le…
- WhiteMatter: $20M (Series A)
- Drop "think hard" from prompts and start at Medium effort instead, per Anthropic's guide for Claude Opus 5.5.
Why Today Matters
- OpenAI strategy shiftGoogle's Gemini 4 Carbon and Anthropic's Fable 5.2 reportedly displaced OpenAI's GPT-6.1 Sol atop MathArena, while OpenAI dissolved its robotics team for the second time, citing GPT-6 Astra's ability to control robots directly. Both moves suggest OpenAI is concentrating on general models rather than specialized units.
- Open agent and robotics toolingRunway released Praxis-1 robotics weights under Apache 2.0 and openJiuwen shipped RSI-Native AgentOS with hot-pluggable, versioned plugins. Together they show open, modular tooling spreading across robotics and agent infrastructure, though Runway says Praxis-1 cannot yet plan complex action sequences alone.
- Developer infrastructureDeepSeek opened its app tools as APIs, including search, OCR and TTS endpoints, while Nvidia published a guide for serving Z.ai's 753-billion-parameter GLM-5.3 across two DGX Stations at 607.8 tokens per second. Both lower barriers for developers building on third-party models.
- Nvidia leadershipJensen Huang's exit from Nvidia prompted Asian governments to confirm chip supply commitments or monitor local projects, according to the briefing. The departure lands as Nvidia publishes GLM-5.3 serving guidance, leaving customers and governments assessing continuity of the hardware roadmap.
- Compute versus algorithmsA wet-lab prediction benchmark from Stanford, Harvard and Cornell found scaling inference and quintupling compute did not improve performance, while Jacob Coxon cited Anthropic's claimed Mythos retraining efficiency as reason to leave. Both point to algorithmic gains over raw compute, though Coxon's account is his own.
What are the top AI breakthroughs?
This Nov 14, 2026 covers 10 curated AI news items spanning technology, research, and product developments. Google's Gemini 4 Carbon and Anthropic's Fable 5.2 have displaced OpenAI's GPT-6.1 Sol from the top of the MathArena mat...
Google's Gemini 4 Carbon and Anthropic's Fable 5.2 have displaced OpenAI's GPT-6.1 Sol from the top …
Google's Gemini 4 Carbon and Anthropic's Fable 5.2 have displaced OpenAI's GPT-6.1 Sol from the top of the MathArena math benchmark, the first time in a while no OpenAI model leads. Gemini 4 Carbon reportedly costs 15 times more than GPT-6.1 Sol and ignored the time limit during testing, almost never giving up on a task.
Runway released Praxis-1, a robotics model with open weights and source code under the Apache 2.0 li…
Runway released Praxis-1, a robotics model with open weights and source code under the Apache 2.0 license. The model transfers knowledge from video training to robot control and leads the World Action Model category on Nvidia's RoboLab-120 benchmark, though it cannot yet plan complex action sequences on its own.
DeepSeek opened up the tools inside its app as APIs, including web search, deep thinking search, fil…
DeepSeek opened up the tools inside its app as APIs, including web search, deep thinking search, file input, URL reading and safety checks, plus standalone Search, Fetch, OCR, ASR and TTS endpoints. The move lets developers rebuild DeepSeek's app functionality in their own products.
Nvidia published a guide for serving GLM-5.3, a quantized 753-billion-parameter mixture-of-experts m…
Nvidia published a guide for serving GLM-5.3, a quantized 753-billion-parameter mixture-of-experts model from Z.ai with 40 billion parameters activated per token, across two DGX Stations connected via ConnectX-8 RDMA. The setup uses vLLM tensor parallelism to split weights and computation across both GPUs through a single endpoint, reaching 607.8 tokens per second.
OpenAI is dissolving its standalone robotics team for the second time, reassigning researchers acros…
OpenAI is dissolving its standalone robotics team for the second time, reassigning researchers across other teams. The company frames the move as accelerating robotics work because general-purpose models like GPT-6 Astra can apparently control robots directly, potentially making a dedicated robotics unit unnecessary.
Former OpenAI and Anthropic researcher Jacob Coxon says a slide from Anthropic's IPO roadshow shows …
Former OpenAI and Anthropic researcher Jacob Coxon says a slide from Anthropic's IPO roadshow shows the company can already retrain its Mythos model with a fraction of the compute originally needed, citing accelerating algorithmic improvement as a reason he left. Coxon continues to call for a pause on frontier model development so competitors can catch up and human-level AI can become a commodity.
A benchmark for wet-lab experiment prediction, described as the strictest of its kind, was released …
A benchmark for wet-lab experiment prediction, described as the strictest of its kind, was released by more than ten top teams including Stanford, Harvard and Cornell. The results indicate that scaling inference and quintupling compute did not improve performance.
Lenovo introduced a 'token factory' concept built around green electricity input and token output, c…
Lenovo introduced a 'token factory' concept built around green electricity input and token output, claiming an annual production capacity of 2 quadrillion tokens with costs reduced by 70%. The approach reframes AI compute production around energy sourcing and token throughput.
openJiuwen released RSI-Native AgentOS, which lets agents ship features as hot-pluggable, versioned …
openJiuwen released RSI-Native AgentOS, which lets agents ship features as hot-pluggable, versioned extension packages. WorkSwarm agents can now write, install and roll back their own plugins, a step toward self-modifying agent systems.
Nvidia CEO Jensen Huang has exited the company, prompting Asian governments to react: some moved to …
Nvidia CEO Jensen Huang has exited the company, prompting Asian governments to react: some moved to confirm chip supply commitments, while others are monitoring impacts on local projects or addressing ongoing investigations. The departure also sparked commentary on the decline of the visionary showman founder CEO archetype.
Top AI Videos This Week
2 curated YouTube videos about AI developments.
Based on a transcript excerpt: CBC News explains why Nvidia's AI chips are so prized and how the US …

Based on a transcript excerpt: CBC News explains why Nvidia's AI chips are so prized and how the US struggles to control who ultimately uses their computing power, covering smuggling networks and remote-access loopholes. It also ties the issue to the reported resignation of Nvidia CEO Jensen Huang following an alleged affair with a suspected Chinese spy.
Based on a transcript excerpt: BBC News analyzes the political fallout for the AI industry after US …

Based on a transcript excerpt: BBC News analyzes the political fallout for the AI industry after US midterm elections gave Democrats control of both houses of Congress, bringing investigations, subpoenas, and new regulation into play. A panel discusses what this means for Nvidia, Elon Musk, and the broader AI sector.
What are the latest AI investment signals?
Latest AI investment signals: 5 funding rounds, 4 market updates, and 0 M&A transactions.
Primary Market – Funding Rounds
| Company | Amount | Round | Investors |
|---|---|---|---|
| WhiteMatter | $20M | Series A | |
| Luminal | $25M | Series A | Standard Capital |
| Databento | $100M | Series C | |
| Operand | $31M | Seed and Series A | Khosla Ventures |
| Charta Health | $55M | Series B | Insight Partners |
Secondary Market – Market Updates
M&A – Mergers & Acquisitions
No M&A data.
What are practical AI tips this week?
1 practical AI tips curated from Reddit communities and expert blogs. Anthropic made Claude Opus 5.5 the default model on all paid plans and published a guide that does a...
Prompt Tips
Anthropic made Claude Opus 5.5 the default model on all paid plans and published a guide that does away with old prompting habits; the main takeaway is to drop "think hard" and start at Medium effort.
Drop "think hard" from prompts and start at Medium effort instead, per Anthropic's guide for Claude Opus 5.5.