搜索“Agent”

🚀 Hy3 is now live in WorkBuddy global—free for users worldwide through August 31, 2026 (PT). Research, analyze data, create documents and presentations, and handle workplace communications in an agentic AI workspace—no setup required. Hy3 is also available through Tencent Design Miora and Tencent Cloud TokenHub. TencentHy Hy3 WorkBuddy Miora TencentCloud

匹配字段正文

Alongside the models and workflows, the project releases training data and evaluation methods intended to help the broader research community build and study open agentic systems. Artificial intelligence is rapidly moving beyond static question-answering toward autonomous agents that can plan, reason, and act across complex, multistep environments. These systems can fix bugs in complex codebases, navigate the web on a user’s behalf, and manage workflows involving calendars and email. While there

匹配字段正文

friendly and experienced engineering friend - no eval experience needed. Tl;dr Ask a question, get an answer (with proof) To use Ori Eval, you can simply tell your favorite coding agent: Your agent hands the request to Ori Eval. Ori Eval explores your codebase, and it comes back with questions. Before it runs any evaluations, it works with you to help figure out what’s important to you — is it cost, performance, latency, speed, tool call accuracy, etc. It then chooses the 5 latest models that fi

匹配字段正文

ting the model since Sakana Chat first launched, and this version improves on earlier ones both in the quality of its Japanese responses and in its ability to carry out work as an agent. For the technical background at the time of the Sakana Chat release, see this blog. Why we built it Ever since Sakana Chat launched, we have heard the same request many times over: people wanted to use this model from their own products and workflows through an API. At the same time, the choice of LLM APIs that

匹配字段正文

e testing Anthropic took more than two months to alert cops; AI systems are fully capable of carrying out nightmare attacks against infrastructure and nobody's ready Meanwhile, AI agents move robotic arms, 'make OT device operator screens lie'; Nvidia found $1B under the couch to help secure American scientific computing dominance Commitment comes as GPUzilla prepares to fortify Uncle Sam's arsenal with at least seven AI-optimized supers; TSMC taps GlobalFoundries to bolster US silicon interpose

匹配字段正文

y 26% and delivers up to 2.5x greater token efficiency; MAI-Voice-2-Flash now powers Dynamics 365 Contact Center, where customers like T-Mobile and EasyJet build their call center agents, reducing GPU costs by up to 89%; MAI-Transcribe-1.5 now serves Dragon Copilot’s multilingual workflow across 58 languages — a solution used by 170,000 medical providers that processed 28 million patient encounters last quarter, where our tests show a 50% relative in reduction transcription and language-identifi

匹配字段正文

ter if based on advanced heuristics or a small model that reads each turn and picks which LLM to use, a router will always be less capable than the model it’s choosing for. Replit Agent lets the model decide instead. The main agent, or core loop, chooses its subagents’ tier and effort, and adjusts its own as the task unfolds. Given that freedom, GPT-6 Astra hands routine implementation to less costly subagents and decides for itself where its tokens are worth spending. On both DeepSWE and Termin

匹配字段正文

re their conclusions agree and diverge. According to Perplexity comms chief Jesse Dwyer, Model Council functions as a skill that tells an orchestrator model to spin up parallel subagents, each acting as a full agent harness and tackling the problem independently. “A wait barrier holds until every perspective finishes, then the model the user selected acts as the ‘chair,’” Dwyer told us in an email. “It reads all the perspective outputs and synthesizes them, explicitly surfacing where the models

匹配字段正文

Announcing Mastra Factory From issue to production, run by agents.Sam Bhagwat· Jul 27, 2026· 2 min read Software production is becoming a hierarchy of loops. Engineers move “down” into lower-level loops to debug reliability. They move “up” to design better processes. The highest-value human work is deciding which loops to create. So today we’re launching Mastra Factory, a system for agents to take software from issue into production. In Mastra Factory, specialized agents triage issues, write and

匹配字段正文

t rather than per token, so there is no position axis to shard, and the KDA layers stay TP-sharded by head. The whole feature is one flag, --dcp-size N. What this buys shows up on agentic traffic, where hundred-thousand-token sessions pile up in the cache. Replaying real coding-agent sessions on 2×4 GB300, with a host-memory KV tier on both arms and DCP as the only difference: DCP removes the active-set wall. The host tier rescues re-prefill traffic, but at 16 concurrent sessions the active work

匹配字段正文

level intelligence, the top open model in the world, at a fraction of closed-model cost. It is 1 at frontend code, strong at writing, and it reliably finishes the long, multi-step agentic tasks where other models stall. Michele Catasta, the President of Replit, told us, "Kimi K3 on Replit Design exceeded our expectations. Unprecedented product design and UI capabilities, at a fraction of the cost we've come to expect from frontier models." Silas Alberti, SVP of Research at Cognition, highlighted

匹配字段正文

Announcing OpenWorker! An open-source agent that doesn't just chat with you, but delivers finished work -- like hand you a polished document, send a slack message, or update a calendar entry. Ask it to prepare a customer brief, untangle your calendar, draft a report, or triage a Slack alert. It works across your files and everyday tools, produces the deliverable, and checks in before doing anything consequential. OpenWorker runs on your Mac, with Windows support coming soon. It does not lock you

匹配字段正文

we introduced our hill–climbing machine, our integrated data, model, and harness flywheel. Today we are excited to share two examples inside Microsoft: MAI models specialized for agentic workloads in GitHub Copilot and Excel. Early results are promising. In our live product deployment, we see that our MAI model deployed in Excel is on par with GPT-5.6 for the most common tasks while being more cost-efficient. The figure below shows how MAI-Code-1-Flash, post-trained within the GitHub Copilot ha

匹配字段正文

odels on quality, run at a fraction of the cost, and are fully customizable for your task. That combination has made them the foundation for teams building serious AI products and agents. But the strategic motivation for adopting open-weight models is still the ability to exercise control. Unlike their closed alternatives, open-weight models give teams control over performance, quality, and functionality. Many commercial AI applications have adopted a majority open-weight endpoint strategy so th

匹配字段正文

Grok Build can now run workflows : describe a large task in plain language, and Grok plans it, fans it out across hundreds of parallel agents in the background, and reports back when everything is done. Your session stays free the whole time. When to use a workflow Workflows are built for complex, multi-faceted tasks a single conversation can't hold: reviewing every feature in a large PR, triaging the last 100 issues, auditing a codebase for one class of bug. If the work splits into many indepen

匹配字段正文

The open research ecosystem has produced many great datasets for the three main agentic domains - software engineering, terminal use, and web research - but every one of them ships with its own harness, its own image conventions, its own grading scripts, and its own failure modes. 23 tasksets behind one taskset API We integrated them all. We ship first-party integrations for 23 agentic tasksets behind one taskset API, unified runtime/sandbox hooks, and one command: 198,000 software engineering t

匹配字段正文

lled tools, everything. Resume it later and there's no re-cloning the repo or reinstalling dependencies. One command sets up your outpost. Standing up secure infrastructure for an agent to run on your own machines usually means real work: a queue watcher, credential handling, sandboxing, lifecycle management. uvx modal-devin init does all of it, and generates code you own and can keep customizing: a scheduler function (the orchestrator from above) and a session function (the worker): Each team g

匹配字段正文

Today we're bringing Grok into Microsoft Outlook. The add-in puts an agent beside your inbox: it summarizes long threads, drafts replies in your voice, and keeps your mailbox organized. Legal came back on the renewal terms Five stories to start your day A lunar landing, a marathon record, and more We need sign-off from finance and legal All 1,204 checks green on main Three flows to review, deck attached Current rate holds if we commit by Jun 30 Accruals due to finance by Wednesday Q3 Vendor Cont

匹配字段正文

t, Ryzen and data center GPUs with up to 2× faster with 70% less VRAM and no accuracy loss via our custom Triton kernels and math algorithms. We also support optimized ROCm builds for GGUF & Safetensors inference. Unsloth is an open-source local UI for faster LLM training and inference, with tool-call healing, code execution, secure web search, remote APIs, and HTTPS deployment. Connect local models to Claude Code, Codex agents and run the latest Kimi, GLM, DeepSeek, Qwen3.6, and Gemma 4 models.

匹配字段正文

At Modal, we build sandboxes, among other things. Agents run in sandboxes, and agents are eating software. Today, Modal runs millions of sandboxes per day, supports up to fifty thousand concurrent sandboxes per customer, and supports a variety of use cases at scale, from reinforcement learning to background agents. Increasingly, our users require more and more sandboxes, created at higher and higher rates. Reinforcement learning can require running millions of sandboxes concurrently, and creatin

匹配字段正文