Cainew

Kwaipilot/KAT-Coder-V2.5-Dev is a code generation model designed for practical programming tasks, building upon the KAT-Coder architecture. It represents an updated version with improvements for coding assistance and development workflows.

ProCreations/grug-27b is a 27 billion parameter language model offering an open-weight alternative for general-purpose AI tasks. The model aims to provide accessible capabilities for various natural language processing and reasoning applications.

Turn your business data into AI workflows and custom apps. Connect any system, migrate any data, and build anything that fits your business 100%.

Stop splitting team knowledge between Slack and everyone's private AI chats. PromptQL is multiplayer AI for teams: think Claude or ChatGPT in shared threads. Tag teammates to review, correct, and refine answers without losing the reasoning. Connect databases, SaaS apps, coding agents, and events. PromptQL captures tribal knowledge and suggests shared-brain updates so context compounds. Scopes and multi-user permissions keep the right context accessible to the right people.

Tell Codi what you want to build. It writes the plan, runs AI agents in parallel across all your projects, and picks the cheapest model that can do each task, so you ship more and spend less. When you add a project it writes the charter and task list itself, and only pulls you in when a decision needs you, through one inbox.

Now in Beta: Public Sites! Create auto-generated, public-facing help centers and knowledgebases at https://moxiedocs.app. Features rich layouts, white-labelling, full search, and no vendor lock-in. Integrates directly with Changelog! Example: https://docs.moxiedocs.app Also shipped: Webhooks & Slack integration for indexing/PR alerts UI/UX pagination & bulk action improvements DB/infrastructure upgrades for faster loading Continued improvements on document generation quality

Wispro turns your voice into writing, instantly. Just talk, messy or unfiltered, and Wispro pastes clean, ready-to-use text directly into whatever you're working on. It adapts to how you think, not just how you speak. Basic Mode captures every word exactly as said. Smart Mode cuts filler words and rambling, formatting it into polished prose. Command Mode turns a spoken instruction into finished writing, emails, replies, whole drafts, on the spot. One voice. Three ways to write.

Quaso turns your busywork into automation across all of your apps and the open web. Simply prompt what needs to get done and it builds and sets it up for you. Connects to 3000+ integrations (Gmail, Slack, Linear, Stripe, etc.) and picks up a real browser when there's no native API. Runs on schedules, triggers, and on demand. Fork templates or save any thread as your own reusable skill. Tag @Quaso in Slack so your whole team can put it to work. Just relax and get stuff done.

OpenCode Superapp brings the agentic power of Codex to the models and infrastructure you choose. Run cloud, local, or self-hosted models in a native workspace where agents understand your projects, work with files, Git, and terminals, speak with you through voice, and operate Mac apps through supervised Computer Use. Private by design and extensible through skills and MCPs, it gives you capable agents without giving up model choice or control. Built from the ground up with Codex and GPT-5.6.

Cosyra gives you a persistent workspace for cloud agents. Your repos, branches, files, dependencies, terminals, and agent work stay together. Open it from your phone and keep building without leaving a laptop awake.

Basedash is the AI analytics platform — chat with an AI data analyst, automatic daily insights, automations, and dashboards. Now all of it is available through the API. Power your product's customer-facing analytics with a fully custom UI, or extend your internal BI in ways no tool anticipated. Send a question, stream the answer, render the chart — every query scoped to the right customer. Ranked #1 on BI Bench and powered by GPT-5.6. The AI data analyst, now yours to build on.

Basement is now a shopping browser. Today we launch agentic checkout: your AI shopper finds the best price across the web and buys it for you. Every purchase runs on a single-use card scoped to one order, so your real card is never exposed and nothing goes beyond the budget you set. Every payment lands in one place. Powered by Crossmint. It knows what you like. Now it can buy it. Live on iOS and Chrome.

Health in ChatGPT now lets eligible U.S. users securely connect medical records and Apple Health to get more personalized insights and better understand their health.

Framework has introduced a new Desktop option equipped with AMD Ryzen AI Max+ Pro 495 processor and 192GB of memory for enhanced performance. The new configuration targets users seeking high-powered computing capabilities with advanced AI processing features.

Reinforcement learning with verifiable rewards has become the predominant recipe for eliciting test-time scaling in explicit Chain-of-Thought reasoners. Yet this scaling path remains computationally costly, since every intermediate step must be decoded as a language token. Latent reasoning instead carries intermediate computation as continuous vectors and already matches or surpasses explicit CoT at far shorter horizons. Despite this promise, latent reasoners remain largely imitation-bound, whil...

As large language models and AI agents become the primary consumers of search results, document set quality determines the upper bound of downstream generation. Yet existing evaluation systems remain confined to scoring documents independently and aggregating via nDCG, ignoring inter-document interactions (redundancy, conflict, complementarity) and unable to answer what makes one document set better than another. To address these issues, we propose a complete evaluate-diagnose-optimize framework...

3D Gaussian Splatting (3DGS) achieves high-quality novel-view synthesis by optimizing freely placed primitives in 3D and adaptively densifying them in under-reconstructed regions. However, this scene-adaptive capacity allocation is largely lost in existing feed-forward 3DGS methods, which commonly regress Gaussians at input pixels and lift them along camera rays. Such pixel-aligned formulations make the number and placement of primitives depend on image resolution and input viewpoints rather tha...

Reinforcement learning with verifiable rewards (RLVR) has substantially improved language-model reasoning, yet its extension to vision-language models remains constrained by the lack of training data that are simultaneously broad, exactly verifiable, and reproducible. We introduce Trace, a taxonomy-guided environment for multidomain visual reasoning. Trace factorizes task construction into a scene grammar and an executable task program, separating visual realization from answer computation. A sh...

Full-parameter post-training of trillion-parameter-scale MoE models introduces substantial system-level challenges for large-scale distributed training, including severe memory pressure, non-overlapped communication overhead, and inefficient kernel execution. While most large-scale LLM training systems are built around GPU-based clusters, this report presents an end-to-end optimization practice on the Ascend NPU SuperPOD. Using the DeepSeek-V4 model family as the target workload, we develop a hi...

Large language models can answer scientific questions, yet a correct output does not reveal whether the model represents or uses the governing physics. Here we show that materials science mechanism information in the open-weight google/gemma-4-E4B-it model has three experimentally separable forms: concepts are readable in individual hidden states, constitutive orientation is carried by controlled transformations between states, and selected internal representations causally control engineering a...

This work introduces G-MAD, an open-source framework that uses Arma3 to generate synchronized multi-view RGB-T data for aerial object detection. G-MAD addresses key limitations of real-world aerial dataset construction, including limited viewpoint control, imperfect RGB-T alignment and high annotation cost. The framework supports structured scenario specification, controllable multi-view camera placement, simultaneous visible/thermal capture, and automatic bounding box annotation using engine-le...

Natural-language autoencoders score explanations of hidden activations by reconstruction: an explanation is deemed faithful if the activation can be regenerated from it. The test is structurally insensitive to individual false claims: if flipping a claim does not change the reconstruction, the claim is never penalized. We show the test is passed in two ways, neither faithful. On a released Qwen-2.5-7B verbalizer, explanations reconstruct well above chance while ~2% of specific claims are reconst...

As autonomous agents rapidly evolve, their ability to reliably manipulate ubiquitous digital documents has become critical for enabling general-purpose AI assistants and automating complex workspace workflows. In this paper, we introduce DocOps, a deterministically verifiable evaluation framework underpinned by a hierarchical taxonomy that deconstructs document operations inspired by real-world practices into atomic dimensions and escalating workflow complexities. Based on DocOps, we systematica...

Recent autoregressive video diffusion methods are increasingly built upon Self Forcing, where the student is trained on histories produced by its own rollout rather than ground-truth video contexts. This reduces exposure bias, but the historical key-value cache is still used by future frames only as frozen rollout state. As a result, future losses cannot supervise how earlier generated latents should be written into more useful keys and values for later video-latent generation. We call this the ...

Contextual entrainment is the tendency of a model to let auxiliary context in its input pull its output, independently of whether that context is relevant, true, or even meaningful. Recently, it has been identified and given a mechanistic account in unimodal language models. Whether and how it manifests in vision-language models (VLMs) is, by contrast, largely unexamined, and the field lacks a purpose-built instrument with which to investigate it. We take the position that studying contextual en...

Practical robotic grasping in complex scenes requires both 3D spatial reasoning and alignment with task-specific requirements. Vision-language models (VLMs) offer a natural way to specify these requirements using language, but existing approaches either use a VLM to predict the grasp directly with limited spatial awareness, or train the VLM together with the grasping model, which requires significantly more data and compute. These limitations impede performance and have prevented scaling to mult...

Current text-to-SQL benchmarks fail to adequately address the complexity and challenges present in real-world database systems. Developers argue that new benchmarking methodologies must better reflect actual data store difficulties to accurately evaluate AI performance.

OpenAI and Anthropic are joining forces to address concerns about the risks posed by open-weight AI models to their commercial interests. The collaboration reflects how leading AI companies are consolidating their efforts to influence policy and protect their market position.

DARPA and the U.S. Air Force successfully demonstrated autonomous AI control of an F-16 fighter jet in a collaborative flight test. The achievement marks a significant milestone in military AI development and autonomous aircraft capabilities.

U.S. startup founders are calling on the government to avoid restricting access to open-weight Chinese AI models, arguing against potential regulatory barriers. They contend that blocking these models could hinder innovation and competitiveness in the domestic AI industry.

Alphabet's increasing cash burn for AI infrastructure and research is alarming Wall Street as Big Tech companies escalate their spending to compete in the AI race. The financial pressure raises questions about profitability and return on investment for major tech corporations.

The open-source software community faces challenges protecting free and open-source software (FLOSS) commons from being exploited or misused by large language models. The article examines strategies to preserve the integrity and sustainability of the FLOSS ecosystem.