What happened in AI today

1. Amazon's TorchNeuron lets PyTorch code run on AWS Trainium chips

   Amazon · Products

   Amazon says TorchNeuron lets developers run their existing PyTorch code on AWS Trainium chips with a single configuration change and no rewrite. In a side-by-side test, Amazon engineers ran the same AI training script on a conventional chip and on Trainium, and the model learned the same things at the same rate. TorchNeuron supports Trainium2 and Trainium3 and will be released under the Apache 2.0 open-source license at the PyTorch Conference North America on October 20.

   • Amazon's TorchNeuron runs PyTorch on AWS Trainium chips

2. MiniMax releases M3.1 Flash, a lower-cost model for agent tasks

   MiniMax · Models

   MiniMax has released M3.1 Flash, a Flash-tier model that it says is strong at agentic tasks, coding, computer use and cybersecurity. The company says the model's usage cost for agentic tasks is 70% lower than its predecessor, and that it runs nearly three times faster in online inference. It is priced at $0.15 per million input tokens and $0.60 per million output tokens, and it is available through the MiniMax API, M Plan and MiniMax Code. MiniMax says the model weights will be released later.

   • MiniMax M3.1 Flash: Redefining Agent Performance and Cost

1. Anthropic commits 5% of its shares to a new public benefit fund

   Anthropic · Business

   Anthropic says it will commit 5% of its shares to the Anthropic Public Benefit Fund, subject to stockholder approval. The fund's stake can grow by an extra 2% at the board's discretion for every 50% rise in company value above a set baseline. Anthropic says the fund will be a wholly-owned LLC, and none of its equity will be used for lobbying or political giving. The company says that, with its co-founders' pledge and an employee equity donation match, roughly 20% of its shares will be dedicated to the public good.

   • Investing in the public benefit

More stories

1. DeepSeek says its in-house search engine powers web and API search

   DeepSeek · Products

   DeepSeek says its web and API search run on its own search engine, which handles hundreds of millions of requests from around the world each day. The company describes the engine as AI-native, indexing hundreds of billions of web pages and supporting multilingual search. DeepSeek says it filters out SEO spam, GEO manipulation, and low-quality AI-generated content, and it points users to its Search & Fetch guide for usage details.

   • Opening up the in-house search API

2. DeepSeek upgrades V4.1 Flash to DeepSeek-V4.1-Flash-1017

   DeepSeek · Models

   DeepSeek says its V4.1 Flash model has been upgraded to DeepSeek-V4.1-Flash-1017, and its official web and app models now run the new version. The API call method and pricing are unchanged, and developers use the deepseek-flash name to reach the latest model. DeepSeek reports gains across benchmarks, including 42.3 on Terminal-Bench 4.0 and 77.0 on DeepSWE v1.1, and says the model also cut task runtime and output length.

   • DeepSeek-V4.1-Flash-1017 Update

More stories

1. Google Research puts AI agents on the wet-lab bench

   Google · Research

   Google Research says it is connecting AI agents to physical lab robots so they can design molecules, run experiments, and learn from the results. Its VERID framework is a live digital twin of the lab workcell that tracks well volumes, pipette tips, and robot arm position, and it simulates protocols before the robot touches reagents. In aptamer tests over three design rounds, the agents raised confirmed on-target hit rates to 33.5% for myoglobin and 42.1% for α-thrombin, the latter at 98.5% specificity. The team also reports that its de novo design pipeline produced binders with single-digit nanomolar affinity for fibroblast activation protein alpha, a cancer-linked target.

   • Grounding AI agents in biological reality: Building blocks for autonomous wet labs

2. Google DeepMind uses Antigravity agent to search for antimicrobial peptides

   Google · Research

   Google DeepMind researchers describe how biologists used Google Antigravity, the company's agentic development platform, to search UniProt for new antimicrobial peptides. The agent queried databases, checked novelty, compared AlphaFold structures and read the genomic context around a candidate bacteriocin from Streptococcus pseudopneumoniae. Google says the candidate has predicted activity against pathogens involved in respiratory tract infections and is now in wet-lab validation, along with several others found through the same pipeline. Google also introduced a Biomolecular Structure Viewer, a 3D tool inside Antigravity that the researcher and agent can both use and annotate.

   • Accelerating the search for antimicrobial peptides with Google Antigravity

3. Mercedes-Benz and NVIDIA extend partnership on automated driving

   NVIDIA · Business

   Mercedes-Benz and NVIDIA announced they will extend their partnership to develop the next generation of driver assistance and automated driving systems. The work combines Mercedes-Benz's MB.OS operating system with NVIDIA's Hyperion compute and sensor architecture, Alpamayo Stack driving software and Halos safety system. NVIDIA says the technology supports point-to-point urban assistance, active safety and automated parking, and the development of future SAE Level 4 applications, including robolimousines. Mercedes-Benz also uses NVIDIA technology for vehicle engineering simulation and in MO360, its digital production ecosystem.

   • Mercedes-Benz and NVIDIA Expand Strategic Partnership to Advance Safe Highly Automated Driving in the Next Decade

4. ElevenLabs adds Instant Soundtracks for scoring and sound effects on uploaded video

   ElevenLabs · Products

   ElevenLabs says its new Instant Soundtracks feature in ElevenCreative generates original music for any uploaded video, then adds sound effects at timestamps it picks, such as a whoosh timed to a logo. Users can also use the video's existing audio as a reference to shape the score's genre, tempo, or instrumentation. The effects appear as separate layers in the Music Video Editor, where users can adjust or swap them. The feature is also available on the ElevenMusic platform and through the ElevenCreative Studio API.

   • Introducing Instant Soundtracks

More stories

1. OpenAI offers free GPT-Rosalind access to U.S. academic labs

   OpenAI · Products

   OpenAI introduced the Rosalind Academic Plan, which gives eligible academic labs up to 30 Business Standard seats at no cost for up to 12 months. Access to GPT-Rosalind, OpenAI's life sciences model, requires trusted-access approval. The plan also includes OpenAI's commercially available frontier models, subject to usage limits, but does not include API access at this time. OpenAI says research data shared through the plan is not used to train its models by default. The plan is open to labs at U.S. universities. PIs at universities without an OpenAI Enterprise agreement can apply through the trusted access process, while labs at universities with an agreement must apply through the university's Enterprise process.

   • Bringing GPT-Rosalind to academic research labs

2. Anthropic releases Code Review for Claude Team and Enterprise

   Anthropic · Products

   Anthropic has moved Code Review, a Claude Code feature that checks pull requests for bugs, out of research preview. It is available to Claude Team and Enterprise organizations. Anthropic says most reviews now finish about 3x faster than in preview, down from about 20 minutes to 6-10 minutes. Admins can set Quick, Standard or Deep review levels for each repository. Eligible Enterprise organizations can claim at least $10,000 in Code Review credits by January 15, 2027, and the credits expire 30 days after claiming. Terms apply.

   • Code Review is now generally available

3. Anthropic gives METR employee-like access for independent risk reviews

   Anthropic · Policy & safety

   Anthropic says it has started giving a six-person team from METR, a nonprofit that evaluates frontier AI models, access roughly on par with its own risk assessment teams. METR will publish its analysis of Anthropic's risk levels, incidents and practices. Anthropic says it cannot veto publication but can require redactions in narrow categories, such as customer data, attorney-client-privileged information and model architecture details. Anthropic also says it is working with SecureBio and Accenture, is in early talks with others including Hugging Face, and is building three tiers of evaluator access, with METR in the third.

   • An update on independent evaluators

4. Anthropic opens Singapore office and signs MOU with Singapore's digital ministry

   Anthropic · Business

   Anthropic opened its fifth Asia-Pacific office in Singapore and signed a Memorandum of Understanding with the Ministry of Digital Development and Information. The MOU sets out cooperation on AI safety, security and model evaluation, research on AI adoption, and responsible adoption across the public and private sectors. Anthropic says Singapore ranks second of 121 countries in Claude.ai usage relative to population, and that more than 1,500 GovTech Singapore engineers and technical staff have adopted Claude Code. The company says the office is open and hiring.

   • Anthropic opens Singapore office and signs MOU with the Government of Singapore

5. Circle to Search on Android can now check text messages for scams

   Google · Products

   Google says Circle to Search on Android can review suspicious text messages to help users spot scams before they click a link or share personal information. Users long-press the home button or navigation bar, then draw a circle around the message. An AI Overview appears with guidance and suggested next steps. If the results show red flags, Google suggests deleting the message and blocking the number.

   • Check if a text message is a scam using Circle to Search.

6. Google expands Android earthquake alerts to more than 200 regions

   Google · Products

   Google is expanding its Android Earthquake Alerts System from 98 countries and regions to more than 200, including Panama and Guatemala. The company says it has sent over one billion alerts since 2021. Its "Be Ready" alert will gain a map of detected shaking, estimated magnitude, suggested actions and a learn-more button in the coming months. Google also published a Crisis Resilience white paper. It says its flood forecasts reach more than two billion people in 150 countries, WeatherNext can predict cyclones up to 15 days out, and fire boundary maps are live in 34 countries.

   • Android earthquake alerts now available in more than 200 regions
   • How we're helping the world prepare for natural disasters

7. Google says 400 publications now take part in its News AI pilot

   Google · Business

   Google says more than 400 publications worldwide now take part in its News AI pilot program, which it launched last year. New partners include Axios, CNN, The Financial Times, Forbes, Post and Courier, and Texas Tribune. Google pays participating publishers for the delivery and use of enhanced content to develop AI features. These include article overviews on Google News publisher pages and customizable audio briefings.

   • 400 publications are now partnering with us on our News AI pilot program.

8. Google adds more links and source context to AI Mode and AI Overviews

   Google · Products

   Google is updating its AI search features with more inline links, shown higher in responses more often. It will test a section at the top of a response that points to official pages, with context on why each link is useful. Carousels of timely articles, which highlight a user's Preferred Sources, are rolling out to more locations and languages. Google says use of Preferred Sources has grown more than 7x since the start of the year, and it started testing clickable creator names on video results in AI Mode this week.

   • More ways to connect with the web from Search

9. Google launches Universal Cart in the U.S. for Search and Gemini

   Google · Products

   Google has launched Universal Cart in the U.S. across Google Search and the Gemini app. The cart saves products from multiple retailers in one place, including items added in AI Mode in Search and in Gemini. Google says it uses the Shopping Graph's 60+ billion product listings to send alerts about price drops and items back in stock. It also uses Google Wallet to compare card perks and shows promo codes in the cart.

   • An intelligent new shopping cart

10. Meta and EssilorLuxottica start AI glasses art tour pilot at Milan's Pinacoteca di Brera

    Meta · Products

    Meta, EssilorLuxottica and Milan's Pinacoteca di Brera have launched a pilot called "Sensing Art" that uses AI glasses to make the museum more accessible to blind and low-vision visitors. The glasses connect to the museum's Grande Brera app and its DescriVedendo audio tour, which turns an artwork's shapes, composition and details into spoken words. At one of eight selected artworks, a visitor can start artwork recognition from the glasses and hear the description through the open-ear speakers. Visitors can also ask Meta AI questions or reach sighted volunteers through Be My Eyes. The tour is available by reservation, with glasses offered Tuesday through Friday from 10:00 AM to 4:30 PM.

    • Meta, Milan’s Pinacoteca di Brera, and EssilorLuxottica Expand Access to Art Through AI Glasses

11. Microsoft AI finishes pre-training MAI-Base-2, an 80B-active, 2.8T-parameter model

    Microsoft · Models

    Microsoft AI says it has completed pre-training MAI-Base-2, a mixture-of-experts foundation model with 80 billion active and 2.8 trillion total parameters. The company says it outperforms every open-weight model it tested on held-out evaluations and is 4-12x more pre-training FLOPs-efficient than DeepSeek-V4-Pro-Base, the strongest open base model. Microsoft says it trained the model on 16k GB200 GPUs over 38 days in August and September 2026, with 21% average MFU and 89% goodput. The model is now in post-training, and Microsoft says it plans to license the weights to trusted partners through Foundry, citing organic training data that does not involve distillation from other models.

    • Frontier Pre-training at Microsoft AI

12. Microsoft Research unveils Magnetar, a faster zero-knowledge virtual machine

    Microsoft · Research

    Microsoft Research has described Magnetar, a zero-knowledge virtual machine that proves programs written in Rust, Go or C. A zero-knowledge proof lets anyone check in milliseconds that a program ran correctly, without rerunning it or seeing its data. Microsoft says Magnetar is 10 to 39 times faster on average than the fastest existing zkVMs across 15 benchmarks, and its proofs verify in about 12 milliseconds. Its security relies only on standard hash functions such as SHA-256 or SHA-3, with no trusted setup.

    • Proof-based general-purpose computing with Magnetar

13. Volvo Cars expands partnership with NVIDIA on automated driving

    NVIDIA · Business

    Volvo Cars says it is expanding its partnership with NVIDIA to build automated driving technology for future Volvo cars. The system will use NVIDIA's Hyperion sensor and compute architecture, Halos OS and Alpamayo Stack driving software. Volvo says it will debut in a next-generation Volvo car as part of its recently announced largest-ever product push, with a gradual introduction of eyes-off and mind-off driving. The companies have worked together since 2017.

    • Volvo Cars and NVIDIA Expand Partnership to Advance Safe Automated Driving

14. Adobe and NVIDIA expand partnership to speed up Photoshop, Premiere and other tools

    NVIDIA · Business

    Adobe and NVIDIA announced a wider collaboration to optimize Photoshop, Premiere, Substance 3D, Frame.io and Brand Concierge. Adobe says Photoshop, Premiere and Substance 3D are optimized for the NVIDIA RTX Spark superchip, which launches Oct. 16, and deliver up to 2x faster AI processing, editing, color and effects. Early results from selected Frame.io workloads show about 5x faster embedding generation, 4-5x better P95 processing latency and an approximately 80% smaller search proxy. Brand Concierge is now optimized with NVIDIA Nemotron open models and will reach enterprise customers later this year, and NVIDIA has picked it for customer interactions on NVIDIA.com.

    • Adobe and NVIDIA Expand Collaboration to Speed Agentic Creation

15. NVIDIA joins European Commission initiative to widen scientists' access to AI tools and compute

    NVIDIA · Business

    At the AI in Science Summit in Dublin, the European Commission launched the AI in Science Enablers Initiative, linked to RAISE, and NVIDIA joined it. NVIDIA says it will offer open models and software, a free open source courseware hub from its Deep Learning Institute, and compute access for startups and small businesses through European AI factories in France, Spain and Sweden. NVIDIA says it has released more than 250 open models, 250 open datasets and over 1,400 open source software projects. The company also says 35 supercomputers are in development across Europe, which it says will serve over 3 million researchers.

    • NVIDIA Expands Advanced Technology Access for Europe's Scientific Community Through Open Learning Resources, Collaborations and Compute

16. Amazon launches Bee Frontier wearable and Bee Eclipse phone attachment

    Amazon · Products

    Amazon introduced two new Bee personal AI devices that turn conversations into summaries, to-dos and reminders. Bee Frontier is a wristband or clip-on wearable with IP67 water resistance and up to seven days of battery life, and it is available now for $74.99. Bee Eclipse is a credit-card-sized device that attaches magnetically to a phone, with more than 21 days of battery life, and it will be available soon from $99.99. Amazon says audio is processed in real time and deleted, and Bee can now connect to Alexa+ and Amazon.com. Those two integrations, along with third-party app integrations, need Bee Pro at $11.99 per month, and new customers get 30 days free.

    • Bee Frontier and Bee Eclipse: Introducing new AI wearable and phone attachment

17. Mistral releases Timestral, a family of time-series forecasting models

    Mistral AI · Models

    Mistral has released Timestral, its first family of time-series foundation models, in four sizes from 5M to 209M parameters. Mistral says the largest model ranks first among pre-trained, zero-shot models on the GIFT-Eval, TIME and fev-bench benchmarks while using about two-thirds of the parameters of the previous best. The models produce probabilistic, multivariate forecasts, accept past and known-future covariates, and run on a standard server CPU without a GPU. Mistral lists the license as modified MIT, has published a GitHub repository and a Hugging Face collection, and offers fine-tuning on private data through Mistral Forge. A technical report is still to come.

    • Introducing Timestral

18. NVIDIA and AI4Bharat announce 2.8 trillion token Indic pretraining dataset

    Hugging Face · Open source

    NVIDIA and AI4Bharat, the Indian language research lab at IIT Madras, announced Nemotron-pretraining-Indic-v1, a corpus of about 2.8 trillion tokens covering the 22 languages listed in India's Constitution. It has native-script data for all 22 languages and romanized text for twelve. The release is due soon, along with the open, GPU-accelerated pipeline that made it, which includes the prompts, LLM workflows and quality checks. The data is translated from curated English sources in Nemotron-CC-v2, and NVIDIA says it complements naturally written Indic text rather than replacing it. NVIDIA says 10 billion Indic tokens drawn from the corpus and natural text raised scores on five Indic benchmarks for Nemotron 3.5 Super VL by 4.98 to 9.82 percentage points. That result covers the whole training stage, not the synthetic data alone.

    • Scaling Indic Pretraining: 2.8 Trillion Tokens and an Open Data Generation Pipeline

19. IBM releases Granite Switch 4.1 with decision and speech adapters in one model

    Hugging Face · Models

    IBM's Granite Switch 4.1 models, 8B and 30B, combine chat, decision-making and speech-to-text in one model served from a single endpoint. They include a new Decision aLoRA for general decisions, a Guardian aLoRA for safety decisions, and Granite-Speech-5.0-Turbo-CTC, which IBM calls the fastest ASR model on the OpenASR Leaderboard. IBM says the combined model uses about the same GPU VRAM as the base model alone, and that adapters can reuse the base model's KV cache. IBM also says the Guardian adapter outperforms several named models on average on safety-related decision tasks.

    • Granite Switch 4.1: Chat, Decisions, and Speech from One Endpoint

20. Swiss AI releases Apertus 1.5 open models with image and audio input

    Hugging Face · Open source

    Swiss AI has released Apertus 1.5, an update to its 8B and 70B Apertus 1.0 models. The new version accepts images and audio alongside text, has an optional thinking mode, and has a 262,144-token context window, four times larger than Apertus 1.0. The weights were released to Hugging Face on July 24, 2026 under an Apache 2.0 license, and the team says audio input is still experimental. Swiss AI says Proton Lumo and Infomaniak Euria have integrated the models, and it has announced a roadmap for Apertus 2.0 with a sparse/MoE design and about 600B total parameters.

    • Apertus 1.5 Release Notes

21. GitHub makes agentic autofix for code scanning alerts generally available

    GitHub · Products

    GitHub says agentic autofix is now generally available for GitHub Code Security and GitHub Advanced Security customers on GitHub Enterprise Cloud, including the data residency option. When a code scanning alert is assigned to Copilot, it explores relevant files, proposes a fix, reruns CodeQL to check that the alert closes, and opens a draft pull request for review. Users can assign up to 25 alerts at once for a single pull request, or assign alerts through the REST API. It works only on alerts on the default branch. An enterprise administrator must turn on the new "Agentic autofix" policy, and a Copilot billing method must be set. The feature uses AI Credits from the Code Security budget and GitHub Actions minutes.

    • Agentic autofix for code scanning alerts general availability

22. Cognition launches Devin Oncall to investigate production alerts

    Cognition · Products

    Cognition's Devin has launched Devin Oncall, which investigates alerts and bug reports and helps teams diagnose and resolve production incidents. It is generally available. A responder watches a stream of alerts or bug reports, diagnoses issues automatically, and remembers underlying problems so it does not repeat investigations. Devin joins the incident channel, investigates alongside engineers, and keeps a live incident document. The company says Devin Oncall scored 100% end-to-end success on the SREGym benchmark, with the fastest time to mitigation.

    • Introducing Devin Oncall

23. Black Forest Labs releases FLUX 3 Image Klein, a 5.6B open weights image model

    Black Forest Labs · Models

    Black Forest Labs released FLUX 3 Image Klein, a 5.6B parameter open weights model for text-to-image generation, image editing and multi-reference generation with up to 10 reference images. The company says it ranks first among the open weights models in its internal comparison, ahead of larger models such as Qwen-Image 2.1 and Krea 2, and gains 91 Elo in text-to-image over FLUX.2 [klein] 9B. The INT8 build runs on NVIDIA GeForce RTX GPUs with as little as 8GB of VRAM, and the weights are on Hugging Face and in the BFL API, with API pricing starting at $0.009 per 1K image. The FLUX Kommunity License allows free commercial use for individuals and companies with annual revenue under $5M, plus research and non-commercial use. Larger organizations need self-hosted licensing or the API.

    • FLUX 3 Image Klein

24. Cohere signs letter of intent to acquire security startup AISLE

    Cohere · Business

    Cohere announced a proposed agreement to acquire AISLE, a company that builds autonomous systems to find and fix unknown vulnerabilities in critical software. Cohere says the deal would add security capabilities to its sovereign AI platform and follows its acquisitions of Reliant AI and Aleph Alpha. AISLE was founded in 2024. The company says it found more than 500 previously unknown vulnerabilities with CVEs across more than 100 codebases in its first year, including all 12 in OpenSSL's January security release and 18 of the 24 CVEs in curl's latest release. The deal is a proposal, and the article gives no price or closing date.

    • Cohere and AISLE sign letter of intent to advance cyber-resilient sovereign AI