Anthropic released Claude Opus 4.7, featuring a significant jump in visual reasoning (from 69.1% to 82.1%) and improved coding performance on benchmarks like SWE-bench Pro (53.4% to 64.3%). While priced the same as 4.6, a new tokenizer can increase token usage by up to 35%, potentially leading to faster cap limits for users.
Sources: Anthropic News, Hacker News, Keep Bookmarks, Latent Space, Simon Willison, TLDR Newsletter, The Neuron
Google has introduced its new Gemini 3.1 Flash TTS model, a text-to-speech tool designed to enhance AI-powered audio generation. This model features advanced capabilities like per-speaker voice control, inline audio tags for emotion and pacing, and support for over 70 languages, offering developers granular control over speech output. It is currently available in preview via the Gemini API and Google AI Studio.
Sources: DeepMind Blog, Simon Willison, The Neuron
OpenAI has introduced GPT-Rosalind, its inaugural life-sciences AI model, designed to enhance scientific research in areas like biochemistry, genomics, and drug discovery. This frontier reasoning model aims to accelerate complex scientific processes and is currently in research preview with partners including Moderna, Amgen, and Thermo Fisher.
OpenAI overhauled its coding app, Codex, into a comprehensive agent workstation that offers Mac-level computer use, an in-app browser, persistent memory, automations, and over 90 new plugins including Atlassian Rovo and Microsoft Suite. This positions Codex as a direct competitor to agentic coding tools like Claude Cowork.
OpenAI's GPT-5.4 Pro has reportedly solved a 60-year-old mathematical conjecture, Erdős problem #1196, with a three-page 'Book Proof' that bypasses previous human approaches, marking a significant advancement in AI's mathematical reasoning capabilities.
Hugging Face has significantly updated its Sentence Transformers Python library, enabling advanced training and finetuning of multimodal embedding and reranker models across text, images, audio, and video. New features like the Router module, CachedMultipleNegativesRankingLoss, and MatryoshkaLoss optimize model flexibility, efficient large-scale training, and deployment, demonstrated by achieving state-of-the-art performance in visual document retrieval.
OpenAI has enhanced its Agents SDK, separating the agent harness for open-source customization and enabling execution delegation to partner sandboxes, fostering long-running, durable agents with advanced primitives. This move makes 'Codex-style' agents more reproducible and immediately attracted integrations from major players like Cloudflare, Modal, and Vercel.
Researchers have developed CROP (Cost-Regularized Optimization of Prompts), an automatic prompt optimization method that significantly reduces token consumption in Large Language Models by 80.6% while maintaining high accuracy. This innovation provides a cost-effective solution for deploying efficient agentic AI systems in production environments.
Google is launching a comprehensive suite of new AI-powered features across its core products, including Google Maps, Search, and Translate, all driven by Gemini models. These updates introduce conversational travel planning with 'Ask Maps' and a 'Canvas' tool for custom itineraries, agentic restaurant booking, advanced live translation via headphones, and expanded AI store calling for shopping assistance, aiming to streamline complex tasks and offer personalized, automated experiences.
Amazon Bedrock Guardrails has introduced Automated Reasoning checks, a new feature that uses formal verification to mathematically validate AI outputs against defined rules, ensuring provably correct and auditable results. This innovation is being rapidly adopted by organizations like Lucid Motors for finance forecasting, First Education & Technology Group for student safety, and Amazon Logistics for EV charging point design reviews. The feature significantly enhances compliance, reduces review times, and addresses the limitations of probabilistic AI across highly regulated industries.
Amazon deployed an AI agent that automatically cancels user accounts, leading to creators losing years of order history, digital libraries, and income without warning or appeal. This highlights concerns about AI's role in due process and the potential for significant user impact when moderation is fully outsourced to LLMs.
Mike Krieger, Anthropic's CPO, resigned from Figma's board on the same day reports emerged that Anthropic is developing and planning to ship its own design software, causing Figma's stock to slide. This signals Anthropic's potential expansion into new application areas and increased competition in the design software market.
Canva rebranded itself as 'an AI platform with design tools' during its $42 billion IPO test, emphasizing its new AI 2.0 features that allow users to describe projects in plain English for editable designs with persistent memory and orchestration. This strategic shift positions Canva as a major AI player in the creative space.
Factory raised $150 million from Khosla Ventures at a $1.5 billion valuation to develop autonomous coding agents that dynamically switch between AI models based on task complexity. This significant funding round intensifies the competition in the rapidly growing market for AI coding agents.
The White House is reportedly preparing to grant federal agencies access to Anthropic's Mythos model, while Google is in negotiations with the Pentagon for a classified deal involving its Gemini AI. These developments indicate a growing integration of advanced AI models into government operations and national security.