Models
New AI model called "Count Anything" does exactly what it says, and that's harder than it sounds
A new AI model called 'Count Anything' can accurately count various objects in images. This capability enhances image analysis and could improve applications in inventory management and visual data processing.
Google Research's Gemini-SQL2 tops text-to-SQL benchmarks by a wide margin
Google Research just launched Gemini-SQL2, which outperforms competitors in text-to-SQL benchmarks. This means developers can expect more accurate and efficient SQL query generation from natural language inputs.
Microsoft's SkillOpt boosts GPT-5.5 by using nothing but a trained Markdown file
Microsoft just boosted GPT-5.5 with SkillOpt, leveraging a trained Markdown file for enhanced performance. This means users can expect more efficient and accurate outputs from the model.
Claude Fable 5 outpaces GPT-5.5 by 13 points on FrontierMath's toughest problems
Claude Fable 5 just outperformed GPT-5.5 by 13 points on FrontierMath's toughest problems. This shows Claude's growing strength in tackling complex mathematical challenges, which could enhance its utility in educational and professional settings.
Google DeepMind releases DiffusionGemma, a model that runs local AI 4x faster
Google DeepMind just released DiffusionGemma, a model that runs local AI four times faster. This boost in speed enhances performance for users running AI tasks on their own devices.
Google's new open model DiffusionGemma generates text from noise instead of word by word
Google just launched DiffusionGemma, an open model that generates text from noise instead of building it word by word. This approach could enhance creativity and efficiency in text generation tasks.
Claude Fable won’t answer basic biology questions
Anthropic's Claude Fable struggles with basic biology questions, failing to provide accurate answers. This limits its effectiveness for users seeking reliable information in that subject area.
[AINews] Anthropic Claude Fable 5 — Mythos but Safe, with Controversial Terms
Anthropic just launched Claude Fable 5, focusing on safety while introducing some controversial terms. This update aims to enhance user experience while navigating complex ethical considerations in AI interactions.
Initial impressions of Claude Fable 5
Anthropic just released Claude Fable 5, showcasing improved conversational abilities and better context handling. Users can expect more coherent and contextually relevant interactions with this update.
llm 0.32a3
Simon Willison just released LLM 0.32a3, an updated version of his language model. This version includes performance improvements and better handling of specific tasks, making it more efficient for developers.
Anthropic releases Claude Fable 5 and Mythos 5 with major gains in coding and science
Anthropic just released Claude Fable 5 and Mythos 5, showcasing major improvements in coding and scientific tasks. These updates enhance the models' capabilities, making them more effective for developers and researchers alike.
New open-source voice model listens nonstop and decides every 0.4 seconds whether to speak or stay silent
Researchers launched a new open-source voice model that listens continuously and decides every 0.4 seconds whether to speak or stay silent. This advancement enhances real-time interaction capabilities in voice applications.
Reality: The Final Eval — Lukas Petersson and Axel Backlund of Andon Labs
Lukas Petersson and Axel Backlund from Andon Labs are launching a new evaluation framework called Reality. This tool aims to enhance the assessment of AI models by providing more accurate performance metrics.
NVIDIA Nemotron 3 Ultra now available on Amazon SageMaker JumpStart
NVIDIA just launched the Nemotron 3 Ultra on Amazon SageMaker JumpStart. Users can now access this advanced model for enhanced machine learning tasks directly within the SageMaker platform.
[AINews] Microsoft Build: MAI-Thinking-1 and MAI Family models
Microsoft just launched the MAI-Thinking-1 model and the MAI Family of models. These updates enhance AI capabilities for reasoning and decision-making tasks, improving user experience in applications that require complex problem-solving.
Microsoft's new MAI models
Microsoft just launched new MAI models designed to enhance AI capabilities across various applications. These models aim to improve performance and efficiency for developers integrating AI into their products.
Ask AI what goes with chicken and the answer depends on whether it learned from recipes or molecules
AI models are now determining food pairings based on either recipes or molecular structures. This means users can get tailored suggestions depending on the learning approach of the AI.
OpenAI gives GPT-5.5 Instant a readability upgrade while phasing out two older models
OpenAI just upgraded GPT-5.5 Instant for better readability and is phasing out two older models. Users will experience improved clarity and engagement in their interactions with the updated model.
9 demos of Gemini Omni and Gemini 3.5 in action
Google just showcased 9 demos of Gemini Omni and Gemini 3.5 in action. These demos highlight the capabilities of the models in various applications, enhancing how users can leverage AI in their workflows.
Claude Opus 4.8: "a modest but tangible improvement"
Anthropic just released Claude Opus 4.8, featuring modest improvements in performance. Users can expect better response quality and more reliable outputs in their interactions.
llm-anthropic 0.25.1
Anthropic just released an update for their language model, version 0.25.1. This update enhances performance and stability, making it more reliable for developers using their API.
Training Azerbaijani language models on Amazon SageMaker AI
AWS just launched training for Azerbaijani language models on Amazon SageMaker AI. This makes it easier for developers to create applications that understand and generate Azerbaijani text.
Anthropic ships Claude Opus 4.8 as a "modest but tangible improvement" that tops GPT-5.5 in most benchmarks
Anthropic just launched Claude Opus 4.8, claiming it outperforms GPT-5.5 in most benchmarks. This update brings modest but tangible improvements, enhancing user experience with better performance metrics.
Microsoft's MAI-Image-2.5 pulls even with Google's Nano Banana 2 on benchmarks
Microsoft just released MAI-Image-2.5, matching Google's Nano Banana 2 on key benchmarks. This competition pushes both companies to enhance their image processing capabilities, benefiting users with improved performance.
🔬ESMFold2: The Bitter Lesson is Coming for Proteins - Alex Rives, BioHub
ESMFold2 just launched, offering a new approach to protein folding predictions. This model aims to enhance accuracy in understanding protein structures, which could accelerate drug discovery and biological research.
Google’s new anything-to-anything AI model is wild
Google just launched a new anything-to-anything AI model that can generate diverse outputs from various inputs. This model expands creative possibilities for developers and users, enabling more versatile applications across different domains.
Towards Speed-of-Light Text Generation with Nemotron-Labs Diffusion Language Models
Hugging Face just introduced Nemotron-Labs, a new diffusion language model aimed at achieving speed-of-light text generation. This model promises to enhance the efficiency and responsiveness of AI-generated content, making it more practical for real-time applications.
Perplexity Is Open-Sourcing Bumblebee
Perplexity is open-sourcing Bumblebee, their new AI model designed for enhanced conversational capabilities. This move allows developers to customize and integrate Bumblebee into their applications, boosting accessibility and innovation in AI interactions.
llm-gemini 0.32
Gemini just released version 0.32 with improved performance and new features. Users can expect faster processing and enhanced capabilities for handling complex tasks.
Gemini 3.5 Flash: more expensive, but Google plan to use it for everything
Google is rolling out Gemini 3.5, which comes with a higher price tag. This move signals their intent to integrate it across all their services, enhancing AI capabilities throughout their ecosystem.