Models & Research

AIAI-generated summary Apple CEO John Ternus announced plans to continue improving and expanding Siri AI following its recent rollout. Ternus states that future developments may include agentic capabilities while focusing on security and privacy protections.

AIAI-generated summary Google previews Gemini 4 Carbon as its strongest frontier model with capabilities for cybersecurity and long-horizon engineering. Google says the model starts testing with a subset of cyber defenders and trusted partners through the Fairwind Program.

AIAI-generated summary Meituan released the LongCat-2.5 multimodal AI model on Monday, which was trained entirely on domestically produced Chinese AI accelerators. Meituan deployed the model inside Miaoshow to help restaurant and hotel operators analyze business performance and execute operational tasks.

AIAI-generated summary OpenAI employee Thibault Sottiaux announced that ChatGPT Dots will use more powerful artificial intelligence models including GPT-6 Astra and GPT-6.1 Terra. OpenAI says these updates will allow ChatGPT Dots to automatically select the appropriate model for specific tasks and provide faster performance.

AIAI-generated summary Nace AI has launched Drex for Finance, a model designed for split-second financial tasks, like determining whether a card payment looks fraudulent. Nace AI says it can use a document or a transaction log along with a fixed set of possible answers, then deliver a probability score for each one.

AIAI-generated summary According to leaks on X, Anthropic and OpenAI are reportedly testing unreleased models including Claude Fable 5.5 and GPT 6.2 Astra. Users observed hidden website code and catalog entries that suggest these updates could arrive soon.

AIAI-generated summary According to community reports, large language models are rumored to directly read and edit the weights of neural networks. This capability allows models to potentially fix bugs and improve performance in a general direction.

AIAI-generated summary Anthropic releases FlashProve, which is an open-source prototype for provable inference using cryptographic zero-knowledge succinct arguments of knowledge. Anthropic says the research aims to accelerate provable inference development and demonstrate that zero-knowledge succinct arguments are practical.

AIAI-generated summary Amazon Web Services, Inc. makes the new Anthropic Haiku 5.5 model available in the Kiro IDE and CLI for the AWS GovCloud (US) Regions. Amazon Web Services, Inc. says developers can use Haiku 5.5 to summarize and compact long context, classify and route requests, process text in bulk, and make small edits across files.

AIAI-generated summary Figma rolled out the Figma agent featuring its first-party editing model for all users. Figma says typical tasks are about 20 percent faster and user ratings are 33 percent higher.

AIAI-generated summary According to a recent paper, separately trained artificial intelligence systems reportedly find a common base of communication under clear conditions. This mathematical proof provides a recipe for achieving the translation process using significantly less data.

AIAI-generated summary The author announced a discussion regarding the construction of a multiplayer game world utilizing autonomous Fable-class models. The publication provides technical insights into building virtual environments with these specific artificial intelligence systems.

AIAI-generated summary OpenAI released GPT-1 in 2018 as a piece of software that users had to download and run on a computer through a terminal interface. The text-based software required technical operation by users to generate basic responses.

AIAI-generated summary The organization officially released the MAI Cyber model inside its designated security harness for users. The organization says the MAI Cyber model has doubled user adoption to 50% in the last 30 days while becoming 20% more token efficient.

AIAI-generated summary NVIDIA releases Nemotron 3.5 Transcribe, an open-weight, promptable speech model now available on Hugging Face and ModelScope. NVIDIA says the model turns recorded speech into useful text by combining transcription, speaker attribution, translation, and audio question answering.

AIAI-generated summary Liquid Foundation Models releases LFM3 as a new generation of multimodal models for text, vision, audio, omni, and decision tasks. The company states that LFM3 models are built for specialized intelligence, allowing fine-tuning on a single GPU and local deployment on various devices.

AIAI-generated summary Baseten released joey-1, a 2B decision model that returns answers directly as a single token and a calibrated probability score. Baseten says the model offers a P50 latency of 20 ms and costs $0.01 per million tokens, making it faster and more affordable than comparable models.

AIAI-generated summary Baseten has released two open-source decision models named mammoth-1 and elephant-1 on Baseten Model APIs. Baseten says mammoth-1 ranks number one on the Jev Decision Index while elephant-1 leads among 9B decision models.

AIAI-generated summary Cartesia releases a blog post exploring raw data processing methods and tokenization tradeoffs in language models. Cartesia says BPE tokenization noticeably improves compute-to-loss efficiency compared to byte-level models.