搜索“Agent”

, legal, and tax work, with every sector weighted by its contribution to U.S. GDP. We see similarly leading performance across other domain specific evaluations, like Vals Finance Agent v2 (multi-step financial research) and Harvey’s Legal Agent Benchmark (legal research and drafting). On AutomationBench, Zapier’s benchmark measuring end-to-end execution across core business functions, Argon ranks 1 with a score of 51.3%. Argon is also uniquely strong when knowledge work requires visual understa

匹配字段正文

53 on the Artificial Analysis Intelligence Index, matching GPT-6 Astra (max, 53) and 1 point ahead of GPT-6.1 Sol (max, 52), with gains driven by lower hallucinations and stronger agentic capabilities. At its current 50% pricing discount and with cache discounts increased to 95%, Gemini 4 Argon costs $1.99 per Intelligence Index task, 60% of GPT-6 Astra (max), but 2.7x GPT-6.1 Sol (max). After the discount ends, this will rise to $3.98 ( 1.2x GPT-6 Astra (max)). Gemini 4 Argon is currently being

匹配字段正文

its × gates) of subroutines that bottleneck important applications. In one example, it beat the published baseline by 40% in a matter of minutes Memory efficiency: A team of Argon agents analyzed fleet-wide profiling telemetry to autonomously identify and apply memory optimizations across Google’s data centers, freeing up over 300 TiB of memory once rolled out, with an estimated 500 TiB to 1 PiB in total savings Large Scale Codebase Migrations and Optimizations: Argon agents are working on migra

匹配字段正文

08 tokens/s as of launch date, faster than GPT-6 Luna (max) at 152 tokens/s. However, it uses 88k output tokens per Intelligence Index task, so it averages 7.1 minutes per task. ➤ Agentic coding is a relative weakness: It scores 1% on Terminal-Bench 4.0 and 22% on AutomationBench-AA. On agentic knowledge work, it scores 1072 Elo on GDPval-AA and 872 Elo on AA-Briefcase, close to Inkling (xhigh). ➤ Low knowledge accuracy, but also relatively high non-hallucination rate: Solar Mini 4 scores -11 on

匹配字段正文

调用和结果检查连接在同一条工作链路中。面对跨网页、文档、表格和代码环境的任务，模型可以根据需要开展资料研究和数据分析，制作 Excel 和 PPT，也可以生成并测试代码。模型会结合中间结果校验判断、调整路径，并根据用户反馈完善交付物。 GDPVal-AA V2.1 结果展示了模型在工作任务中的表现。这类能力也已经进入具体产品形态，蚂蚁集团内部运行的桌面协同 Agent 客户端“小虎”接入了 Ling-3.1-flash。小虎提供桌面工具和执行环境，Ling-3.1-flash 负责理解目标、规划步骤、调用工具、生成代码并整合成果，让模型能够走出聊天框，在桌面环境中完成工作。 🎬 视频演示（见公众号原文 ） 04 医疗沟通：把零散描述转化为可执行的下一步 医疗健康场景的信息往往不完整，也很少使用标准化表达。患者可能只记得几个不适症状，检查报告中包含大量专业指标，治疗和复诊安排还需要结合个人情况持续沟通。 Ling-3.1-flash 将临床理解、专业判断和医患沟通贯穿咨询过程。从澄清症状和梳理病史，到解读检查结果、分析方案选择，再到解释风险和明确下一步行动，模型围绕当前目标持续推进，而不

匹配字段正文

GPT-6.1 Sol is now generally available on Amazon Bedrock, bringing stronger reasoning to coding, computer use, and professional workloads that run frequently. For an AI agent to complete a task, it may need to gather information, use tools, test different approaches, recover from errors, and verify its result. Every decision shapes what happens next. A wrong turn can add model interactions, tool calls, latency, and human intervention before the agent reaches a useful result. The economics of an

匹配字段正文

body of professional output.” See Marty’s workflow Visit The Rundown University . How do you use AI? Tell us for a chance to be featured. 🤖 ZeroDrift - Enforcement runtime for AI agents. Rewrite or block non-compliant output in one API call. Free to start 🗣️ Eleven V4 - ElevenLabs' new voice AI, alongside a faster Turbo version 🖥️ Holo4 - H Company's new computer-use models ⚙️ Manus 2.0 - Upgraded agent platform with video editing and automations Sponsored Listing CData’s ‘The 178x Cost Spread’

匹配字段正文

imization, Flash Attention, and Decision Transformers. A dLLM is a drop-in replacement for a typical autoregressive LLM, supporting all its use cases, including RAG, tool use, and agentic workflows. When prompted with a query, instead of producing the answer one token at a time, the answer is generated in a coarse-to-fine way, as illustrated in the above animation. Improvements are suggested by a neural network – in our case a Transformer model – which is trained on large amounts of data to glob

匹配字段正文

e world's fastest reasoning language model, built to make production AI feel instant. Why speed matters more now Production AI isn't one prompt and one answer anymore. It's loops: agents, retrieval pipelines, and extraction jobs running in the background at volume. In loops, latency doesn’t show up once. It compounds across every step, every user, every retry. Yet current LLMs still share the same bottleneck: autoregressive, sequential decoding. One token at a time, left to right. A new foundati

匹配字段正文

ing a word. Under the dominant autoregressive decoding paradigm, every one of those tokens is generated sequentially, each requiring a full forward pass. On the phone with a voice agent, this generation (decode) time stacks up to the point where you’d rather just hang up. The result is a strange split in the industry. The reasoning intelligence frontier has sprinted ahead, while realtime intelligence has stood still. Voice is one of the only verticals where GPT 5.x, Claude Sonnet, and Gemini Pro

匹配字段正文

are building AI assistants that work alongside software engineers as they edit code, with responsive autocomplete, intelligent tab suggestions, and rapid edit applications. Voice agents that help customers quickly get to what they need. And AI co-pilots that work alongside designers, creatives, and analysts, making suggestions and edits, managing workflows, and doing deep research. All of these solutions operate instantly and in-the-flow. No need to input a query then go pour a cup of coffee wh

匹配字段正文

token context window for processing large documents and maintaining extensive conversations, with native tool calling and structured output support using JSON schemas for building agentic workflows. The API is OpenAI-compatible, making integration with existing codebases seamless. Azure AI Foundry provides enterprise-grade infrastructure, including network isolation, data privacy guarantees ensuring your data stays in your Azure environment and is never used for training, Azure compliance standa

匹配字段正文

for most work, this makes Sonnet 5.5 a strong fit for workloads that run continuously or at scale. In engineering, that includes first response to alerts, always-on monitoring of agents, SQL generation, UI and UX testing, and fast coding agents in the IDE with a fixed spend cap. Across the enterprise, it supports scoped analysis, short spreadsheet edits, and routine document tasks for large user populations. Pairing Sonnet 5.5 with Opus 5.5 to get the job done We recently announced the availabi

匹配字段正文

oding performance jumps sharply The performance gap between Sonnet 5.5 and its predecessor is most significant in coding, according to Anthropic. On Terminal-Bench 4.0, a test for agentic coding, Sonnet 5.5 hits 70.6 percent compared to Sonnet 5's 10.3 percent. On CursorBench 4.0, which recreates real coding sessions from the Cursor editor, Sonnet 5.5 scores 55.5 percent versus 34.1 percent, landing just two points below Opus 5.5 (57.8 percent). On FrontierCode 1.1 at the "High" setting, Sonnet

匹配字段正文

Models built for every interface Holo4 clicks and types on a screen, writes and runs its own code, and calls MCP or API tools. It uses whichever fits the task. Most agentic models are trained for one interface only: GUI-focused models are blind without a screen, while models that prefer tool calling are stuck in front of an application that has no API. Real work is not siloed that way, and a single business task can require combining these different approaches. Holo4 runs on desktops, on the web

匹配字段正文

nnet-5-5；Amazon Bedrock，为 anthropic.claude-sonnet-5-5；AWS 上的 Claude Platform，为 claude-sonnet-5-5；Google Cloud，为 claude-sonnet-5-5；Microsoft Foundry，为 claude-sonnet-5-5，仅限 Global Standard 部署 在 Claude Code 中 从 Claude Code v2.1.284（Agent SDK for TypeScript v0.3.284 或更高版本）起，sonnet 别名在 Claude API 上解析为 Sonnet 5.5。它默认以 medium effort 运行，原生支持 1M 上下文窗口。在 Claude Code 中你无法关闭 Sonnet 5.5 的思考，effort 决定模型的思考量。Sonnet 5.5 没有快速模式。默认模型仍为 Opus 5.5，因此对于范围明确的任务，请使用 /model sonnet 切换。 我们希望你会喜欢试用 Sonnet 5.5，并一如既往地欢迎分享反馈。

匹配字段正文

option today. Fireworks team reports that reasoning models like Kimi K3 sometimes spend more than 90% of generated tokens on internal reasoning. That cost compounds in multi-turn agentic workloads. Each turn replays prior reasoning back to the model. Context grows roughly quadratically with the number of turns. Long traces from early turns get re-read, and re-billed, on every later call. Fireworks team explains how customers wanted K3’s coding capability at lower cost. Turning down K3’s reasoni

匹配字段正文

day tasks — including coding and creating office documents. Sonnet 5, 5.5’s predecessor, was announced about three months ago. At the time, the model’s selling point was efficient agentic deployment — the ability to run agents at a lower cost than competitors. The big selling point with 5.5, meanwhile, is speed. Anthropic claims that Sonnet 5.5 is 30% faster than its predecessor and that its rate of token burn is significantly slower. In the Anthropic hierarchy of models, Sonnet is less powerful

匹配字段正文

DeepSWEの38課題とTerminal-Bench 2.1の30課題。このうちDeepSWEでは、4候補から選ぶ条件で31課題を解決した。 同じ評価でJevに解答を選ばせた場合の解決率は、候補を無作為に1つ選んだ場合を下回ったという。この評価でのCLMの推論速度はJevの4～6倍と報告されており、推論時間の測定にはH100 GPUを使用した。 🧵(6) Agentic Benchmarks We find that Jev fails to serve as a verifier for long-horizon tasks, performing below the random-selection (Pass@1) baseline. In contrast, with lightweight fine-tuning, CLM achieves SOTA performance on challenging agentic benchmarks, including DeepSWE… pic.twitter.com/PElaTjW4vs— Jacky Kwok (@jackyk

匹配字段正文

Introducing Agora-2, our next-generation multi-agent world model. Agora-2 supports up to 20 humans and agents interacting inside a shared environment, all simulated in real time. Our multiplayer research preview is available to try right now! https://t.co/6k0kkrtw2z

匹配字段正文