PulseAugur
EN
LIVE 23:10:50

NVIDIA, Anthropic, Google, and Ideogram release new models and research

By PulseAugur Editorial · [9 sources] ·

NVIDIA has released Nemotron 3 Ultra, an open-weight 550B MoE model with a 1M context window, optimized for long-running agent workloads and boasting significant speed and cost improvements. Anthropic's research suggests current AI systems show early signs of recursive self-improvement, with their internal metrics indicating AI is accelerating AI development. Google released Gemma 4 12B, a multimodal model designed for on-device use with an encoder-free architecture, and Ideogram released its 4.0 image model with open weights, claiming it's the best open image model available. AI

IMPACT New open models and research on AI self-improvement accelerate agent development and push the boundaries of multimodal capabilities.

RANK_REASON Multiple significant open model releases and research papers from major AI labs.

Read on Latent Space (swyx) →

AI-generated summary · Google Gemini · from 9 sources. How we write summaries →

NVIDIA, Anthropic, Google, and Ideogram release new models and research

COVERAGE [9]

  1. Latent Space (swyx) TIER_1 English(EN) ·

    [AINews] not much happened today

    a quiet day of RSI.

  2. Latent Space (swyx) TIER_1 English(EN) ·

    [AINews] not much happened today

    a quiet day

  3. Smol AINews TIER_1 English(EN) ·

    not much happened today

    **Anthropic's Mythos/Opus cycle** sparked mixed reactions with praise for **Claude Mythos**'s one-shot workflows and concerns over **Opus 4.8** benchmark regressions. **Opus 4.7** showed strong chemistry task performance, "making Claude a chemist." **Sakana AI** launched an **RSI…

  4. Smol AINews TIER_1 English(EN) ·

    not much happened today

    **NVIDIA** released **Nemotron 3 Ultra**, a fully open **550B MoE** model with **55B active parameters** and **1M context**, optimized for long-running agent tasks with up to **5x speedup** and **30% cost reduction**. It features hybrid Mamba/attention, LatentMoE, native MTP, and…

  5. Smol AINews TIER_1 English(EN) ·

    not much happened today

    **Microsoft** released the detailed technical report for **MAI-Thinking-1**, a generalist reasoning model trained without third-party distillation, achieving **97% on AIME 2025** and outperforming Sonnet 4.6 in human preference tests. The report was praised for transparency, reve…

  6. Smol AINews TIER_1 English(EN) ·

    not much happened today

    **NVIDIA** led open-source AI model releases with **Cosmos 3**, a comprehensive omnimodal world model unifying language, image, video, audio, and action using a Mixture-of-Transformers design, and **Nemotron 3 Ultra**, a **550B** parameter open-weight model noted for high serving…

  7. Smol AINews TIER_1 English(EN) ·

    not much happened today

    **Anthropic** rolled out **Claude Opus 4.8**, which shows incremental improvements but mixed benchmark results, including better cooperation and coding behavior but some regressions in document parsing. Platform updates include mid-conversation system instructions enhancing long …

  8. Smol AINews TIER_1 English(EN) ·

    not much happened today

    **Harness engineering** is emerging as the key differentiator for coding agents, emphasizing the stack of **model + harness + eval loop** over just stronger base models. **DeepSeek** is building a harness team to optimize interaction and verification loops, while **Google's Gemin…

  9. Smol AINews TIER_1 English(EN) ·

    not much happened today

    **Inference optimization** is increasingly architectural, with **EAGLE 3.1** improving speculative decoding and long-context handling, collaborating with **vLLM** and **TorchSpec**. **Perplexity** open-sourced a rebuilt **Unigram tokenizer** cutting CPU use by **5–6×** and achiev…

RELATED ENTITIES

RELATED TOPICS