marktechpost.com·platform·292 items·last fetched
A technical comparison of Anthropic's Claude Sonnet 5, Sonnet 4.6, and Opus 4.8, focusing on agentic coding benchmarks, API pricing, and cost-performance metrics.
A comparison of Anthropic's Claude Sonnet 5, Sonnet 4.6, and Opus 4.8, focusing on agentic coding benchmarks, API pricing, and cost-performance trade-offs.
Mistral AI has launched Robostral Navigate, an 8B model designed to enable robots to navigate complex environments using only a single RGB camera.
Stanford researchers released TRACE, an open-source, capability-targeted agentic training system designed to improve agent performance by turning recurrent failures into synthetic reinforcement learning environments.
Skyfall AI released MORPHEUS, a persistent enterprise simulation benchmark designed to necessitate continual reinforcement learning in non-stationary environments.
Prime Intellect released Verifiers v1, a tool for decomposing agentic reinforcement learning environments into composable tasksets, harnesses, and runtimes.
Skyfall AI launched MORPHEUS, a persistent enterprise simulation benchmark designed for continual reinforcement learning.
An exploration of 'autoresearch' and 'Bilevel Autoresearch' techniques for developing autonomous ML research loops in AI agents.
A University of Michigan team introduced NeuroVFM, a neuroimaging foundation model trained with Vol-JEPA on over 5 million uncurated clinical MRI and CT volumes.
NeuroVFM is a new neuroimaging foundation model trained with Vol-JEPA on uncurated hospital MRI and CT data to learn brain anatomy and pathology.
Mira Murati's Thinking Machines Lab published research on human-centered AI using customizable model weights.
Ant Group's Robbyant unveiled LingBot-VA 2.0, a causal video-action model specifically designed for physical AI applications.
LingBot-World-Infinity has been introduced as an open causal world model that functions as an interactive, agentic simulator.
Google Research introduced SensorFM, a foundation model for wearable health data pretrained on one trillion minutes of sensor information.
Kyutai released MuScriptor, an open-weight model for transcribing multi-instrument music into MIDI format.
OpenAI Releases GPT-5.6 (Sol, Terra, Luna): A Three-Tier Model Family With Programmatic Tool Calling in the Responses API MarkTechPost
OpenAI has launched GPT-5.6, a three-tier model family (Sol, Terra, Luna) featuring programmatic tool calling via the Responses API.
NVIDIA released Nemotron-Labs-3-Puzzle-75B-A9B, a compressed hybrid MoE LLM designed to significantly increase server throughput while maintaining model accuracy.
NVIDIA released a compressed hybrid MoE LLM, Nemotron-Labs-3-Puzzle-75B-A9B, designed to deliver 2.03x server throughput compared to previous iterations.
Google AI Studio introduced an 'Import from GitHub' feature in Build mode, allowing users to turn existing repositories into editable and deployable applications.
A report comparing Datalab Lift, a 9B schema-first extractor, against other tools like NuExtract3 and LlamaExtract.
Meta Superintelligence Labs introduced Muse Spark 1.1, a multimodal reasoning model tailored for agentic tasks.
Ant Group’s Robbyant released LingBot-VLA 2.0, an open-source 6B vision-language-action model designed for cross-embodiment robot manipulation.
OpenAI Releases GPT-Live and GPT-Live-1 mini: Full-Duplex Voice Models That Delegate Deeper Reasoning to GPT-5.5 MarkTechPost
NVIDIA Releases Audex (Nemotron-Labs-Audex-30B-A3B): A Unified Audio-Text LLM That Preserves the Text Intelligence of Its Backbone MarkTechPost
NVIDIA released Audex, a unified audio-text LLM that integrates audio capabilities while maintaining core text intelligence.
SpaceXAI launched Grok 4.5, a model trained with Cursor specifically for coding and agentic tasks, featuring enhanced token efficiency and competitive pricing.
Ant Group’s Robbyant open-sourced LingBot-Vision, a 1B-parameter, boundary-centric vision foundation model designed to improve dense spatial perception for embodied AI.
OpenAI launched GPT-Live, a full-duplex voice model family for ChatGPT, featuring two versions (GPT-Live-1 and mini) that delegate deep reasoning to GPT-5.5.
NVIDIA released Audex (Nemotron-Labs-Audex-30B-A3B), a unified audio-text LLM capable of understanding and generating general audio while maintaining high text intelligence.
Google AI Studio introduced an 'Import from GitHub' feature in its Build mode, allowing developers to turn existing repositories into editable and deployable applications.
OpenAI Releases GPT-Realtime-2.1 and GPT-Realtime-2.1-mini for Low-Latency Voice Agents in the API MarkTechPost
OpenAI released GPT-Realtime-2.1 and GPT-Realtime-2.1-mini, designed for low-latency voice agents within the API.
Liquid AI open-sourced Antidoom, a method using Final Token Preference Optimization to reduce repetitive "doom loops" in reasoning models.
Tencent introduced Hy3, an open 295B parameter Mixture-of-Experts (MoE) model that utilizes sparse activation of 21B parameters per token.
Sakana AI launched Sakana Translate, a Japanese–English–Chinese translation tool powered by Namazu with translate, proofread, and ask modes.
Synthetic Sciences released OpenScience, an open-source, model-agnostic AI workbench for scientific research in biology, physics, and chemistry.
An exploration of training Gemma-3 for structured mathematical reasoning using Tunix, GRPO, and LoRA adapters.
Meituan has released LongCat-2.0, a 1.6 trillion parameter MoE language model featuring a native 1 million token context window.
LlamaIndex introduced 'legal-kb', an agentic retrieval system using Index v2 with specialized retrieval and text-processing tools.
An overview guide covering current open-source extraction models for converting structured PDF files to JSON format.
Anthropic Launches Claude Science Beta: A Multi-Agent AI Workbench for Reproducible Genomics, Proteomics, and Cheminformatics Pipelines MarkTechPost
NVIDIA introduced HORIZON, a hands-free agent framework designed for hardware design that treats the process as repository-level code evolution.
NVIDIA AI introduced ASPIRE, a self-improving robotics framework that achieved 31% zero-shot performance on long tasks in LIBERO-Pro.
Anthropic has introduced Claude Science Beta, a multi-agent AI workbench designed for reproducible workflows in genomics, proteomics, and cheminformatics.
NVIDIA introduced ASPIRE, a self-improving robotics framework that achieved 31% zero-shot success on LIBERO-Pro long-horizon tasks.
Interfaze released diffusion-gemma-asr-small, an open-source diffusion-based automatic speech recognition model capable of transcribing six languages using a parallel denoising decoder.
Mistral AI released Leanstral 1.5, an Apache-2.0 licensed Lean 4 code agent model that successfully solved 587 out of 672 problems on the PutnamBench benchmark.
Interfaze released diffusion-gemma-asr-small, an open-source diffusion-based automatic speech recognition (ASR) model capable of transcribing six languages.
WebBrain is a new open-source, local-first AI browser agent for Chrome and Firefox that automates tasks by reading pages, with support for local models like llama.cpp and Ollama.