Subscribe

The Day in AI

Monday 16 November 2026

←

In short

  • NVIDIA says Vera Rubin NVL72 delivered up to 2x the per-GPU performance of GB300 NVL72 in its MLPerf Training v6.1 debut.
  • Hugging Face now routes language model requests from Inference Providers through OpenRouter, reaching open-weight models across more than 80 providers with automatic retries.
  • OpenAI says GPT-6 Astra led the detailed proof in Nathaniel Craig's preprint toward defining the Standard Model on a lattice.
  • OpenAI released a free image classifier for vetted organizations that scores potential CSAM, with a separate score for AI-generated material.
  • US Commerce Department reportedly moved to cut China's remote access to AI chips through overseas data centres.

The day in brief

NVIDIA reported on 16 November that its Vera Rubin NVL72 system delivered up to 2x the per-GPU training performance of GB300 NVL72 in MLPerf Training v6.1.

MLPerf is the benchmark suite run by MLCommons, where vendors submit the time it takes to train or serve fixed reference models. This round is the first for Vera Rubin, and NVIDIA says it won every Training v6.1 benchmark. On DeepSeek-V3, NVIDIA reports about 97% scaling efficiency going from 256 to 1,024 GPUs, which cut training time from 16 minutes to about four.

NVIDIA also made the first submission to MLPerf Endpoints, a test of models served to users. Its provisional Kimi K3 result shows more than 10x the throughput per watt of GB300 NVL72 at 50 tokens per second per user. MLCommons has not yet verified that number. At SC26 on the same day, NVIDIA announced Rubin HPC, a simulation GPU with more than 50,000 CUDA cores, expected in 2028.

Hugging Face switched Inference Providers for language models from its own router to OpenRouter. OpenRouter picks a provider for each call by live price, speed and reliability, and retries on another provider if one fails. OpenRouter reports 46% fewer capacity errors under the new routing. Billing is unchanged, with the provider's price passed through at no markup, but custom provider keys no longer work for language models. OpenRouter also reports that open-weight models took 71.7% of its tokens at the end of September 2026, up from 26.8% in October 2025.

OpenAI says GPT-6 Astra led the detailed proof behind physicist Nathaniel Craig's October preprint, which builds a gauge-invariant measure for lattice chiral fermions, a long-open piece of defining the Standard Model on a lattice. Craig says he supplied the strategy and Astra turned it into a construction. The paper reports that 16 Lean 4 modules compiled without errors. Lean 4 is a proof assistant in which a computer checks each logical step. The source check on this record is partial, and it's not yet known how physicists outside the paper judge the construction.

Also in the news

  • OpenAI says its new CSAM classifier flags 48 percent fewer images than its earlier approach, and ChatGPT now pauses accounts seeking abusive material for 24 hours.
  • US Commerce Department released a rule on remote access to AI chips, which reports describe as closing China's overseas data centre route.
  • Alibaba added 10 selectable models to Qwen Office, including Qwen4-Max, which it opened in preview on 11 November with a million-token context.
  • Sarvam AI launched Token Factory, an inference service in India that serves Sarvam 105B and open-weight models such as Kimi K3 and DeepSeek V4 Pro.
  • OpenAI is expanding ChatGPT for Academic Researchers to 200,000 researchers, with 50,000 grants of free ChatGPT Plus opening on 23 November.
  • OpenAI says it signed Apollo Research, METR, Transluce and Goodfire for its first phase of embedded model assessments.
  • Anthropic reportedly cut its coordinated disclosure window for high and critical severity bugs from 90 to 30 days, citing AI-driven bug discovery.
  • Google Threat Intelligence Group reports what it calls the first observed attackers using an autonomous AI framework to turn security patches into exploits.
  • DeepSeek users report a daily chat quota and an input cap near 40,000 Chinese characters, which DeepSeek has not confirmed.
  • Moonshot AI is reportedly steering its Kimi models toward science as coding benchmark gains flatten, according to the South China Morning Post.
  • Fractile, a UK chip startup, reportedly raised $720 million in a Series C.
  • LangChain reportedly raised $300 million at a $3 billion valuation.

Everything from this day

Launches and products

16 Nov 2026NVIDIAVera Rubin NVL72 MLPerf Training v6.1 and MLPerf Endpoints results NVIDIA's Vera Rubin platform debuts on MLPerf, with NVIDIA winning every Training v6.1 benchmark and claiming over 10x throughput per watt versus GB300 on MLPerf Endpoints.16 Nov 2026NVIDIARubin HPC NVIDIA announced Rubin HPC at SC26, a new GPU in the Vera Rubin platform for simulation and data processing, with more than 50,000 CUDA cores on a single GPU.16 Nov 2026Huawei CloudCloudRobo-WM1.2 Huawei Cloud released CloudRobo-WM1.2, a world action model for robots built for its Ascend platform, and says it will open-source it.16 Nov 2026Alibaba (Qwen)Qwen Office multi-model selection (10 models, incl. Qwen4-Max) Qwen Office added 10 selectable models, including the new Qwen4-Max and third-party models such as Kimi-K3.1-Pro, GLM-5.3 and DeepSeek-V4-Pro, so users can match model cost to task.16 Nov 2026OpenAIOpenAI CSAM Image Classifier (omni-moderation-csam-latest) OpenAI released an image classifier that detects potential CSAM, including AI-generated material, free to vetted organizations through the API, with new ChatGPT warnings and 24-hour pauses.16 Nov 2026OpenAIComputer-use agents in ChatGPT for Healthcare OpenAI added computer-use agents, improved clinical answers and new healthcare plugins to ChatGPT for Healthcare, so it can work across healthcare software.16 Nov 2026OpenAIChatGPT for Academic Researchers expansion OpenAI is expanding ChatGPT for Academic Researchers to reach 200,000 researchers worldwide, with 50,000 new grants of free ChatGPT Plus opening November 23.16 Nov 2026Sarvam AIToken Factory Sarvam AI launched Token Factory, an inference service that serves its own models and global open-weight models from GPU clusters in India through one OpenAI-compatible API.16 Nov 2026Hugging FaceInference Providers routing through OpenRouter Hugging Face switched Inference Providers for language models from its own router to OpenRouter, so one Hugging Face token reaches open-weight models across 80+ providers with automatic retries.