NAll news

All news

Saturday203 items

  2. Krea AIFirst-party
    AI score42

    Krea plugin enables turning sketches into cartoon animations within Grok

    Krea announced that with the Krea plugin, users can turn sketches into cartoon animations directly in Grok simply by making a request to Grok. Previously, Krea had launched a plugin for the Grok bot that can generate images and videos within Grok.

    Quoted postKrea · @krea_ai

    introducing Krea for Grok bot. generate images and video with the best models in the world, right from Grok. try it now 👇 https://t.co/jjCDGfLLUY

    Same story · 2 reports

    #Product updates#Image generation#Video

  4. Guillermo Rauch
    AI score25

    Vercel CEO amazed by fx v0.0.16's pace of improvement

    Vercel Developers released fx v0.0.16, saying it has better prompt caching and smaller download packages and binaries: 41% fewer uncached input tokens, about 60% smaller libfx npm download size, and 2.8% smaller CLI binary. Vercel CEO Guillermo Rauch reposted it and marveled at the astonishing pace of improvement, saying it is now smaller, faster, and more token-efficient, that the team is continuously running benchmarks, and that it is required to improve its own performance on both open-source and closed-source models.

    Quoted postVercel Developers · @vercel_dev

    fx v0.0.16 This version has better prompt caching and smaller downloads & binaries. • 41% fewer uncached input tokens • About 60% smaller libfx npm downloads • 2.8% smaller CLI binary Changelog → https://t.co/9n2qtJfh4E https://t.co/GjeEEDT8ze

    #English text is being prepared.#Agents

  5. Artificial AnalysisAnalysis
    AI score55

    Hyperflow H3 ranks 5th in image-to-video and 3rd in video editing in AA video benchmarks

    Artificial Analysis's evaluation shows that Video Rebirth's Hyperflow H3 ranks 5th on AA-Video-I2V v2.1 and 3rd on AA-Video-Editing v1.1, the latter ahead of its base model MiniMax H3 (768p). Hyperflow H3 is an 8-step LoRA for MiniMax H3, released in September 2026 with open weights, and currently supports image-to-video and video editing. Its API pricing is $9.00 per minute for image-to-video and $18.00 per minute for video editing, both at 1088p, less than half the price of Dreamina Seedance 2.5.

    #Evaluations / benchmarks#Video#Open-source ecosystem

  6. OpenAI: Deployment Safety and System CardFirst-partyModel releases
    AI score72

    OpenAI releases the GPT-Rosalind-6 system card addendum, offering restricted access for biological research

    OpenAI releases the GPT-Rosalind-6 system card addendum. This model uses the same set of weights as GPT-6 Astra, with no new training data or post-training added, and is provided only through trusted access to vetted research institutions, enterprises, and government partners. At the highest reasoning budget, MedChemBench v7 averages 49.6% (GPT-Rosalind-5.5: 28.9%), GeneBench Pro v13 single-attempt success rate is 37.7% (previous generation: 16.7%), and LifeSciBench Gold average score is 61.3% (previous generation: 44.0%). Under the Preparedness Framework, this release is rated High capability in the biological and chemical domains, below the Critical threshold; rated Critical in the cybersecurity domain; and below High in AI self-improvement.

    #Safety / alignment#OpenAI#English text is being prepared.

  7. Figma AI / Agent / MakeFirst-party
    AI score45

    Figma Weave in Detail: Turning Creative Processes into Reusable Workflows

    Figma Weave integrates AI generation and editing tools into reusable workflows based on a node canvas, and adds a new way of working in parallel with Figma frames. Workflows support branching exploration and comparing results from the same prompt across multiple AI models, with built-in editing nodes for color correction, compositing, levels, cropping, trimming, and painting. Weave tools are now open for community publishing, so teammates can run packaged workflows directly without manipulating nodes.

    A “Spring Arrivals” campaign design shown in Figma Design beside its connected Figma node in Weave, with options to update the Figma design or review an update.

    #Product updates#Agents#Multimodal

  8. Baseten Engineering BlogFirst-partyAnalysis
    AI score52

    Baseten cuts 770GB incremental weight sync from 36 seconds to 15 seconds

    Baseten's engineering blog states that incremental weight synchronization for the approximately 770GB GLM-5.3 checkpoint dropped from an average of about 36 seconds to 15 seconds, and request pauses fell from 6 seconds to 0.8 seconds. The new design uses the weights on the GPU as the parent copy, no longer keeps a full copy in CPU memory, and uses Adler-32 checksums to perform constant-time updates on changed bytes. The freed host memory is used for KV cache offloading; at a concurrency of 48, the cache hit rate rose from 2% to 92%, and per-GPU output throughput increased 3.2x. For Kimi K3, the 8.4GB incremental sync takes 75 seconds with a 4.1-second pause.

    The fastest delta weight syncs in the world

    #Tutorials / practice#English text is being prepared.#Reasoning

  9. cnBeta
    AI score75

    Google's first data center satellite goes online, writes haiku in orbit using Gemma 4

    After about three weeks of in-orbit testing and calibration, Google's first data center satellite carrying a TPU has entered its formal operational phase. This week, engineers sent prompt instructions to the satellite for the first time, and the pre-installed Google DeepMind Gemma 4 model composed a haiku about space. The satellite can generate up to about 1 kilowatt of power; because the TPU has relatively high power consumption, the chip can only be powered on for about 20 minutes at a time. Google will next evaluate the TPU's reliability under temperature changes and cosmic radiation. If it is verified to be stable, the company plans to launch two more follow-up satellites in 2027 using more advanced chips.

    Same story · 2 reports

    #Industry news#Google#English text is being prepared.

  10. cnBeta
    AI score38

    Starbucks plans to introduce AI personalized menus at drive-thru locations: automatically identifying member identities and recommending drinks

    Starbucks plans to introduce AI personalized menus at its drive-thru locations. When members arrive by car, the system can identify them and, based on their purchase history and preferences, automatically display recommended drinks and simplify the payment process. The company expects to begin testing a back-end AI model in fiscal year 2027 to assist employees in accurately entering orders. It has not yet announced the number of test stores, the AI supplier, or a full rollout timeline. Starbucks management disclosed that about half of customers use the drive-thru service.

    #Industry news#Agents

  11. cnBeta
    AI score50

    AI boom fuels tape storage: Fujifilm invests $64 million to expand LTO-10 tape production

    Fujifilm plans to invest $64 million between fiscal 2026 and fiscal 2028 to expand its Odawara plant in Kanagawa Prefecture, Japan, raising tape production capacity to roughly five times the current level to meet growing demand from cloud service providers, data centers and enterprise customers. The expansion focuses on LTO-10 tape, which offers 40TB native capacity per cartridge and up to 100TB compressed, with a transfer speed of 400MB per second. The massive long-term archiving needs created by generative AI, combined with tape's ability to be physically isolated to reduce ransomware risk, have brought the technology back into the spotlight.

    #Industry news#English text is being prepared.

  12. Testing CatalogProduct release
    AI score58

    Anthropic Renames Claude Max 20x to Ultra and Removes 5-Hour Limit

    Anthropic announced an adjustment to the naming of Claude Max subscription tiers: Max 5x is renamed Max, and Max 20x is renamed Ultra, with prices remaining at $100 per month and $200 per month, respectively. Verified Ultra accounts will have the 5-hour usage limit removed starting today, while new accounts will retain a temporary limit until they complete verification; weekly limits remain unchanged.

    Quoted postClaude · @claudeai

    We’re renaming our Max plans: Max 5x → Max Max 20x → Ultra Max is still $100/month, and Ultra is still $200. If you’re a subscriber, there’s nothing you need to do. We’re also removing the 5-hour limit for verified Ultra accounts, starting today with most existing subscribers. Newer accounts keep a temporary limit until they’re verified, which helps us prevent fraud and abuse. You'll get an email when it's removed. Weekly limits remain unchanged. View our plans here: https://t.co/niRpyGF5Me

    Same story · 5 reports

    #Product updates#Anthropic

  13. UCSD Hao AI LabFirst-party
    AI score48

    Hao AI Lab Open-Sources the JetSpec++ Speculative Decoding Framework

    UCSD Hao AI Lab has open-sourced JetSpec++, which pipelines the draft stage and verification through causal parallel drafting, a hidden-state predictor, and an anchor-token decoder, achieving 1,361 TPS on Qwen3-8B, up to 1.54x end-to-end speedup over DSpark, and 7.45x speedup over autoregressive decoding while remaining lossless. Combined with NVFP4 and kernel optimizations, Qwen3-8B exceeds 4,000 TPS and Qwen3.8-27B exceeds 900 TPS (on two GB200 GPUs). The model, code, and inference engine have all been open-sourced.

    Same story · 5 reports

    #Model releases#Reasoning#Open-source ecosystem

  14. UCSD Hao AI LabFirst-party
    AI score46

    Hao AI Lab Open-Sources the JetSpec++ Speculative Decoding Framework

    UCSD Hao AI Lab has launched JetSpec++, which, through causal parallel drafting and pipelined verification, achieves lossless end-to-end speedups of 1.54× over DSpark and 7.45× over autoregressive decoding on Qwen3-8B, reaching 1,361 TPS in BF16. On GSM8K, Qwen3-8B's per-round latency drops from 9.92 ms with DSpark to 6.84 ms, with drafting time compressed from 3.76 ms to 0.61 ms. The model, code, and inference engine have all been open-sourced.

    Same story · 5 reports

    #Model releases#Reasoning#Open-source ecosystem

  15. UCSD Hao AI LabFirst-partyResearch
    AI score54

    Hao AI Lab proposes JetSpec++: pipelining draft generation and verification, achieving 1.54x speedup on Qwen3-8B

    Hao AI Lab proposes JetSpec++, which pipelines draft generation and verification through causal parallel drafting, a hidden-state predictor, and an anchor-token decoder, achieving lossless speedup on Qwen3-8B and Qwen3.8-27B. Compared with DSpark, it achieves up to 1.54x end-to-end speedup, and 7.45x speedup over autoregressive decoding, reaching 1,361 TPS and 367 TPS respectively under BF16; with NVFP4 and kernel optimizations, it exceeds 4,000 TPS and 900 TPS on two GB200s. Across 7 benchmarks, it accepts 8.94 tokens per round, compared with 6.89 for DSpark; the model, code, and inference engine have all been open-sourced.

    Same story · 5 reports

    #Model releases#Reasoning#Open-source ecosystem

  16. UCSD Hao AI LabFirst-party
    AI score47

    Hao AI Lab open-sources JetSpec++, a speculative decoding method with up to 7.45x speedup

    Hao AI Lab has open-sourced JetSpec++, which pipelines draft generation and verification through causal parallel drafting and a hidden-state predictor, achieving 1,361 TPS and 367 TPS on Qwen3-8B and Qwen3.8-27B respectively—up to 1.54x faster than DSpark and 7.45x faster than autoregressive decoding while remaining lossless. On Qwen3.8-27B, the number of tokens accepted per round improves by 22% over DSpark (6% for 8B); trained with only 16 blocks, it extrapolates to 32 draft blocks, reaching 14.55 (8B) and 14.05 (27B) tokens per round on MATH-500, where DSpark drops to 8.66 and 8.49. Combined with NVFP4 and kernel optimizations, Qwen3-8B exceeds 4,000 TPS and Qwen3.8-27B exceeds 900 TPS on two GB200s.

    Same story · 5 reports

    #Model releases#Reasoning#Open-source ecosystem

  17. UCSD Hao AI LabFirst-partyProduct release
    AI score59

    Hao AI Lab Open-Sources JetSpec++ Models, Code, and Inference Engine

    Hao AI Lab announced that JetSpec++'s models, code, and inference engine are all open-sourced, along with links to the paper, code, and Hugging Face. The method pipelines draft generation and verification, achieving up to 1.54x end-to-end speedup over DSpark and 7.45x over autoregressive decoding on Qwen3-8B, reaching 1,361 TPS in BF16; with NVFP4 and kernel optimizations, it exceeds 4,000 TPS on two GB200s.

    Same story · 5 reports

    #English text is being prepared.#Reasoning#English text is being prepared.

  18. Swyx
    AI score17

    swyx says the AI Engineer conference brings together a thousand engineers

    swyx stated that he wrote about this on AINews: the event is bringing together 1,000 top AI engineers, first to accurately present the SOTA of AI x Code in 2026, and second to preview 2027. He also said he is proud to have convened a group of invited attendees, speakers, and sponsors to echo this year's three major themes.

    Quoted postmichelle · @michellechen

    hiding amongst the greats. see you at https://t.co/HfcmDb4hBz https://t.co/QF6it1JqTR

    #Industry news

  19. Elon Musk
    AI score33

    Musk reposts new Neuralink video

    Sawyer Merritt said Cleo Abram visited Neuralink's Austin headquarters and released a new video, showing its next-generation R1 surgical robot for the first time. Musk reposted it and commented, "Great video about @Neuralink."

    Quoted postSawyer Merritt · @SawyerMerritt

    Cleo Abram (15M followers across all platforms) just released a new video about @neuralink. She visited the company's Austin HQ. "This is Neuralink’s Next-Gen R1 surgical robot. Its main job is to put a computer chip into people's brains, and this is the first time they're unveiling it." https://t.co/wzTuUq2Or0

    #Industry news#Embodied AI

  20. Rohan PaulAnalysis
    AI score59

    Voice Arena launches the Turing TTS benchmark: the best model fools listeners only about 10% of the time

    Voice Arena has launched the Turing TTS benchmark, in which listeners choose the real human among a real human recording and a model-generated recording of the same sentence; the proportion of wrong choices is the Fool Rate, and 50% means indistinguishable. The best-performing Gemini 3.8 Flash TTS made listeners choose wrongly in about 10% of cases. On Voice Arena's preference leaderboard, the top four models are statistically tied, but overall they still show a clear gap relative to real humans.

    Quoted postShobhit Banga · @shobhitbanga

    Drop 2/15 Are we actually measuring the things we ultimately want our voice AI models to achieve? The best AI voices are becoming hard to tell apart. From each other. Not from humans. 🚀 Introducing Turing TTS: an evaluation that measures how often human listeners mistake AI-generated speech for real human speech. We tested leading TTS models against real human speech. Even the best model fooled listeners only ~10% of the time. With Turing TTS, we hope to contribute to a broader discussion within the research community about how we measure progress in voice AI. If one of our goals is to develop speech that is indistinguishable from human speech, how well do our current evaluation methods capture progress toward that goal? We don’t claim that Turing TTS is the right way forward, or that human indistinguishability should be the goal of every TTS system. Rather, we hope these findings encourage the community to collectively examine whether blind pairwise human evaluations of naturalness are sufficient, and what else we might need to measure. Results are now live on @voicearena_ai.

    #Evaluations / benchmarks#Speech#Google