The AI news and updates that matter today, curated from across the AI ecosystem, ranked.
Canva4h agoCanva released Canva Craft, an image generation model developed by Canva Research. It uses a two-stage architecture to plan composition before rendering images at 2K resolution. The model debuted 4th on DesignArena with an Elo score of 1340, ranking first in People and Portraits. Craft is available now in Dream Lab and through Canva’s public API.
1Password15h ago1Password launched 1Password for Agents, a platform allowing personal AI agents to sign in to websites without accessing the user's actual credentials. When an agent requests a login, 1Password autofills the password directly into the site upon user approval. The feature is free and optional for personal and family accounts, with integrations rolling out for Claude, Grok, and others.
Anthropic4h agoAnthropic and Theorem used Claude to formally verify security properties in 123 open-source libraries at the RISC-V assembly level. The process proved that programs will not crash under specific conditions for inputs under 64 MB. While verifying 239 entry points, the model discovered and patched bugs that had eluded years of automated fuzz testing.
Grok Bot4h agoxAI’s Grok Bot now operates within phone-native apps, including Calendar, Reminders, Contacts, Messages, Location, and Home controls. The update rolls out now on iOS and Android, requiring users to update the app and grant per-app permissions to enable these capabilities.
MiniMax4h agoMiniMax released the open weights for M3.1 Flash, a 428B-parameter multimodal mixture-of-experts model. It features a 1M-token context window and achieves 76.2 on BankerToolBench, 73.7% on DeepSWE v1.1, and 82.2 on CyberGym. The model utilizes MiniMax Sparse Attention and DSpark speculative decoding to improve inference efficiency and is now available on Hugging Face and ModelScope.
Liquid AI4h agoLiquid AI launched RIPL, a low-latency inference engine designed to serve Liquid Foundation Models in live request paths. RIPL delivers complete responses in under 50 ms under peak load, with up to 20× faster inference. The engine is currently deployed at Shopify for production search enhancement, achieving sub-20 ms inference on real-world commerce workloads.
Dmitry Lyalin4h agoGoogle released an experimental preview of the Gemini API CLI, a zero-dependency binary for terminal and agent-based workflows. The tool generates images, video, audio, and music, analyzes media, and manages agents directly from the command line. It replaces manual scripts and raw API calls by writing media files to disk and providing structured, machine-readable outputs for autonomous agents.
HeyGen15h agoHeyGen Video 1, the company's first general-purpose video model, ranked first in text-to-video generation in a blind test on Artificial Analysis. The model supports text, image, and reference-to-video modes with resolutions up to 2K and durations between 5 and 15 seconds. It is available via API starting at $0.01 per second during its current launch promotion.
Wenhao Chai4h agoThe FrontierCS team launched FrontierCS 2, a benchmark evaluating AI agents on 48 open-ended computer science research tasks. Agents must engineer solutions that match or beat human reference implementations on hidden workloads within a five-hour budget. On the new Kaggle leaderboard, Claude Opus 5.5 leads with a 58.3% pass rate, while no agent solves 16 of the 48 tasks.
Pruna AI12h agoPruna AI released P-Video-2-Pro-Avatar, a model built on MiniMax H3 that converts a single portrait into a lip-synced talking video. The service costs $0.02 per second of video. A launch promotion offers a 50% discount on usage until November 5 at 18:00 CET.
Artificial Societies4h agoArtificial Societies launched Rothko, a foundation model of human behaviour built to simulate consequential decisions. Unlike standard language models, Rothko uses a proprietary architecture trained on behavioural data to output decision predictions rather than text. The model aims to improve accuracy in market research, AI safety, and swarm alignment by better representing individual belief systems and collective population distributions.
Karan Vaidya4h agoComposio launched Firewall, a runtime boundary that intercepts and filters AI agent tool calls against user-defined rules. It blocks unauthorized actions, data leakage, and prompt injections before requests reach integrated applications like Slack, GitHub, and Gmail. This approach shifts agent security from build-time configuration to active runtime enforcement.
Artificial Analysis4h agoArtificial Analysis launched Coding Agent Index v2.0, incorporating Cognition's FrontierCode v1.1 and revised SWE-Atlas-QnA grading. Claude Code with Claude Opus 5.5 leads the new index at 61.7. The update evaluates models across 290 tasks, with the index now weighted equally across FrontierCode, Terminal-Bench 4.0, and the updated SWE-Atlas-QnA-AA benchmark.
Morph15h agoMorph now hosts an abliterated variant of DeepSeek V4.1 Flash through its Model APIs. Access requires a human-reviewed KYC process for named individuals to ensure defensive use, following recent infrastructure hacks affecting PHP and other brands. The model provides unrestricted capabilities for security research, red teaming, and safety evaluation, with usage logged and subject to revocation.
Agentic AI Foundation4h agoThe Agentic AI Foundation added four hosted projects: Agent Skills from Anthropic, Open Responses from OpenAI, OpenEnv from Hugging Face, and OpenSandbox from Alibaba. These additions bring the foundation to 10 hosted projects in its first year. The projects provide open standards for agent capabilities, model interoperability, reinforcement learning environments, and secure execution sandboxes.
Sakana AI4h agoSakana AI launched Sakana Aquarium, an AI agent platform for organizational decision support and execution. It provides a shared foundation using an ontology to link existing data without migration, managed workflows with human approval, and governance for auditability. The platform packages the groundwork Sakana previously rebuilt for each implementation project, and is available through consultation with Sakana’s forward-deployed engineers.
Perplexity4h agoPerplexity released Decider for Mac, a menu-bar app that runs the 4B pplx-decider-flash-v1.2 model locally. It serves the Decisions API at localhost:8080, enabling text and image classification without an API key or data leaving the machine. The app requires an Apple silicon Mac with macOS 13 or later and 16 GB of memory for the 5.8 GB model.
Faisal Mahmood21h agoFaisal Mahmood’s team released CS1-4B, a 4-billion-parameter decision model trained on 12.4 million questions from 3.6 million patients. The model performs one-pass clinical record analysis on a single desktop GPU. When paired with a generalist LLM, it screens patient populations for clinical trials, achieving higher eligibility accuracy than keyword search while using 690× fewer language-model tokens.
OpenAI15h agoOpenAI hosted two employees from METR and Redwood Research for a three-week on-premises pilot to evaluate its internal system for monitoring agent misalignment. The team provided deep infrastructure access to assess the system's effectiveness. Findings from this pilot helped strengthen existing safeguards and will inform future research collaborations with METR and other third-party evaluators.
Higgsfield15h agoHiggsfield launched Content Factory, a tool that generates 30-second short-form video reels from written ideas. The system produces content like "What if" explainers and ad hooks by applying a selected trending style from a reference gallery to a prompt. Content Factory is available now on the Higgsfield platform, with generation costs set at approximately 31.5 per episode.
vLLM4h agovLLM added day-0 serving support for MiniMax M3.1 Flash, a 428B-parameter mixture-of-experts model. The integration features native NVFP4 KV caching, NVFP4 MoE weights, and DSpark speculative decoding. On NVIDIA Vera Rubin NVL72, the engine reaches 724.9 tokens per second per user decode and 191.5k tokens per second per GPU throughput, reducing checkpoint memory footprint by 44.6 percent.
Supabase12h agoSupabase released an MCP Server block that gives apps their own Model Context Protocol server, exposing app actions as tools for AI agents like ChatGPT or Claude. Every tool call runs as the signed-in user, ensuring existing Row Level Security policies control agent access. Developers install the block via a single command to scaffold the necessary Edge Functions.
Ling21h agoAnt Group announced Ling-3.1-flash-1019, an updated build of its Ling-3.1-flash model. The new version scores 44 on the Artificial Analysis Intelligence Index, up from 41, with agentic performance gains reaching 1524 Elo on AA-Briefcase and 68% on AutomationBench-AA. API access and open weights for the model are coming soon.
SGLang4h agoSGLang added Day-0 serving support for MiniMax's M3.1 Flash model, achieving over 1,000 tokens per second at batch-size-1 decode. This integration, a joint effort with MiniMax, NVIDIA, AMD, and RadixArk, also includes reinforcement learning support via Miles. The model features 428 billion total parameters with 23 billion active parameters per token.
Hugging Face15h agoHugging Face released Carbon-X, an Apache 2.0 licensed 14.8B DNA model built by midtraining Qwen3-14B on 302B tokens. It achieves 66.7% average accuracy across five DNA benchmarks, outperforming Qwen3-14B by 27 points and approaching GPT-6.1 Sol at 1/1,800 the inference cost. Hugging Face released the model openly to ensure researchers can study its capabilities and risks without restrictive guardrails.
Replit4h agoReplit launched Monitoring Agent, an autonomous on-call engineer for apps hosted on the platform. The agent monitors production logs and metrics to detect issues, identifies the root cause, and validates a fix. With auto-publish enabled in Publishing settings, the agent deploys the validated fix directly to production and provides a summary of the changes.
Google DeepMind15h agoGoogle DeepMind, with Google Research, WRI, and IIASA, released Forest Typology (ForTy), a 10-meter resolution map categorizing global land into six forest and land-use types. The release includes an annual 2017–2025 time series tracking land-use changes. Both datasets are available on Google Earth Engine to help organizations build deforestation-free supply chains and protect natural forests.
Gemini12h agoGoogle Gemini now references Google Maps searches and saved places alongside Google Calendar to provide tailored daily recommendations. This integration functions through the Personal Intelligence setting, where Gemini synthesizes location history and schedule data to assist with planning. The feature requires explicit permission to connect these apps to the Gemini experience.
Shadeform12h agoShadeform launched a dedicated CLI and official MCP server to expose its full API for GPU infrastructure. The CLI supports agent mode for structured output in environments like Claude Code and Cursor, while the MCP server enables AI agents to autonomously monitor availability and manage multi-cloud resources, including instances, clusters, and volumes, directly from the terminal or an AI assistant.
Together AI4h agoTogether AI now serves MiniMax M3.1 Flash, a 428B-parameter multimodal Mixture-of-Experts model, as a day-0 launch partner. The model features a 1M-token context window and supports text, image, and video input for agentic, coding, and security workflows. It is available on Together AI serverless and dedicated infrastructure, priced at $0.15 per million input tokens.
Artificial Analysis5OpenAI3Ling2
Claude2SGLang2vLLM2About this page
Keeping up with AI is exhausting — launches, new tools, research, and company moves land every day, scattered across X, Reddit, blogs, and newsletters, and it's easy to miss an update that could impact your work. HeadsUpAI tracks the whole AI ecosystem and surfaces the most significant AI news and updates today. Each one gives you the whole story in a 30-second read, plus the backstory of what led to it and what other companies are doing — presented straight, no hype, no bias — so you keep up with the AI ecosystem without the noise, and act on what matters.
Frequently asked questions