MarkTechPost
PulseAugur coverage of MarkTechPost — every cluster mentioning MarkTechPost across labs, papers, and developer communities, ranked by signal.
23 day(s) with sentiment data
Nace AI and Liquid AI will compete on multimodal decision model pricing and performance
Both Nace AI's Drex Max and Liquid AI's d1-ultra are new multimodal decision models that return probabilities instead of generating text. They have similar benchmark scores and are offered via API. Expect further announcements and comparisons regarding their pricing per token and performance on specific decision-making tasks as they vie for market share.
NVIDIA's open-sourcing of Run:ai will spur adoption of GPU orchestration for AI
NVIDIA's decision to open-source its Run:ai GPU orchestration platform for Kubernetes lowers the barrier to entry for managing complex AI workloads. This move is likely to encourage wider adoption of sophisticated GPU scheduling and resource management techniques within the AI community, especially for those using Kubernetes.
AI agent web-reading costs are a growing concern
TinyFish's emphasis on the overlooked costs of AI agents reading the web, specifically retrieval charges from search and fetch APIs, suggests this is a significant pain point for developers. This highlights a potential market for services that optimize or subsidize these costs, as TinyFish itself offers.
-
Google open-sources Elephant-Goldfish Model for spec-first AI coding
Google has open-sourced the Elephant-Goldfish Model (EGM) plugin, a tool designed to improve AI coding agents by enforcing a specification-first approach. This plugin separates the drafting of a specification by a long-…
-
Inkling 2 Small multimodal MoE model released on Tinker, Apache weights expected
Thinking Machines Lab has released Inkling 2 Small, a multimodal Mixture-of-Experts model, available on the Tinker platform. The model boasts a 594B total parameter size with 17B active parameters and supports a 256K co…
-
Google DeepMind cuts AI video costs with Gemini Omni 1.2 Flash
Google DeepMind has launched Gemini Omni 1.2 Flash, a video generation and editing model that now supports native 15-second clips, an increase from the previous 10-second limit. The update also introduces multi-turn con…
-
Google Research unveils framework for subjective image classification
Google Research has introduced Agile Deliberation, a novel human-in-the-loop framework designed to help users define and build image classifiers for subjective concepts. This framework assists in identifying the boundar…
-
Ricursive Intelligence releases Ripple-65B Preview, a 65B open-weight coding model · 8 sources tracked
Ricursive Intelligence has released Ripple-65B Preview, a 65-billion-parameter open-weight language model. This model achieves a strong performance on the TerminalBench 2.1 coding benchmark, scoring 80.5 and surpassing …
-
New attention methods enable LLMs to process millions of tokens
Researchers are developing new attention mechanisms to handle extremely long contexts in large language models, aiming to overcome the quadratic cost of traditional methods. Papers introduce techniques like Recursive Sp…
-
Nace AI releases Drex Max v1, a 27B multimodal decision model for local inference
Nace AI has released Drex Max v1, a 27-billion parameter multimodal decision model, as open weights. This model is designed to run locally on consumer hardware, including Apple Silicon Macs with as little as 24GB of RAM…
-
Deep Principle releases 753B Hydrogen-1.0 for scientific discovery · 6 sources tracked
Deep Principle has released Hydrogen-1.0, a 753 billion parameter open-weight model designed for scientific problem-solving and research assistance. Built upon GLM-5.3 and fine-tuned on scientific data and laboratory re…
-
Infini-AI-Lab releases Hermit framework; Block unveils Dessa speech model
Infini-AI-Lab has launched Hermit, an open-source framework designed to train AI agents by providing an external layer of instructions, skills, memory, and tools based on task feedback. Concurrently, Block's AI lab has …
-
AI Security Agents Compared: Claude, Codex, Mantis, Cloudflare · 2 sources tracked
Four major tech companies—Anthropic, OpenAI, Cloudflare, and Google—have released AI agents designed to identify security vulnerabilities in source code. These agents share a common methodology: first, an agent detects …
-
Samaya AI's Sterling model beats Claude Fable 5.1 on finance benchmark at lower cost
Samaya AI has launched Sterling, a 3 billion active parameter model designed for financial research. Sterling, built on NVIDIA's Nemotron 3.5 Lightning, outperforms Claude Fable 5.1 on the FrontierFinance benchmark by 3…
-
Infini-AI-Lab releases Hermit, an open-source AI agent adapter framework
Infini-AI-Lab has launched Hermit, an open-source framework designed to adapt AI agents without altering their core models. This framework allows developers to train external 'harness adapters' that can incorporate new …
-
Block unveils Dessa, a speech-only full-duplex AI model
Block's AI research lab has introduced Dessa, a novel full-duplex speech model trained exclusively on human conversation audio. Unlike many current voice agents that rely on text-first architectures, Dessa operates dire…
-
FlashMap boosts LLM decoding speed by 6.14x with novel flow map technique
Researchers from Polymathic AI and Princeton University have introduced FlashMap, a novel method for speculative decoding in large language models. FlashMap utilizes a continuous flow map instead of a discrete diffusion…
-
Liquid AI releases 6 open-weight 2B multimodal models, LFM3
Liquid AI has launched LFM3, a new family of six open-weight, 2B-class multimodal models. These models, built on a shared backbone, are designed for text, vision, audio, and decision-making tasks, with some capable of h…
-
AI adoption in UX shows mixed feelings; Alibaba's Qwen models scale rapidly
A recent survey indicates that while AI tools are accelerating work for user experience practitioners, many have mixed feelings about their impact. Separately, Alibaba's Qwen models have seen significant development, sc…
-
Nace AI launches Drex for Search; Microsoft/Epic's AI stance draws criticism
Nace AI has introduced Drex for Search, a new tool designed to enhance document search capabilities for AI agents. This system reportedly achieves a 92.05% success rate on the BrowseComp-Plus benchmark, utilizing an ave…
-
Anthropic launches free AI scanner for open-source vulnerabilities · 4 sources tracked
Anthropic has released OSS Scanner, a free tool designed to identify vulnerabilities in open-source projects. The scanner utilizes Anthropic's advanced AI models, including Claude Mythos, to analyze codebases and direct…
-
Letta tutorial showcases stateful AI agents with advanced memory features
This tutorial demonstrates how to build a stateful engineering copilot using Letta, an AI framework that extends agentic memory beyond simple prompt-based interactions. The system allows for the creation of inspectable …
-
TinyFish details production architecture for reliable AI agentic search
TinyFish has detailed a production-ready architecture for AI agentic search, addressing common failures in real-world applications. Their approach involves a seven-phase control loop that manages planning, retrieval, ev…