Google 连发 Gemini 4 全系模型
Google 本周密集发布 Gemini 4 系列:Hydrogen(极速编码/多模态)、Carbon(数学推理)、Decision 1(决策模型),并配套第 8 代 TPU。Demis Hassabis 与 Sundar Pichai 亲自站台,Search 已接入 Hydrogen。这是 Google 对 OpenAI 和 Anthropic 的最强反击,定价压至 $0.10/1M tokens。
Google 这次发布节奏明显是刻意设计的「饱和攻击」——一天一个重磅,把舆论周期完全占满。Hydrogen 的 2.5x 速度提升和 Search 深度集成,说明 Google 终于把模型优势转化到核心产品里,而不是像之前 Gemini 那样叫好不叫座。
定价 $0.10/1M tokens 是冲着 Claude 3.5 Sonnet 和 GPT-4o mini 去的,但真正的杀招是 TPU v8 + 模型垂直优化(Carbon 专精代码、Decision 1 专精决策)。如果你在用 AWS/Groq 跑推理,需要重新算一笔 TCO 账:Google 的端到端优化可能让同样预算跑出 3x 吞吐量。
创业团队注意:Google 这次把「agentic work」放在 Hydrogen 的核心卖点,说明他们认为 2024 下半年到 2025 年的关键战场是 Agent 执行层,不是聊天界面。做 Agent 框架的团队,优先适配 Gemini 4 的函数调用和工具使用能力,可能比等 OpenAI 的下一个模型更实际。
RT @OfficialLoganK: Introducing Gemini Decision 1, our new frontier decision model in the Gemini API. It is $0.10 per 1M input tokens wit…
查看原文 →RT @GoogleDeepMind: Meet Gemini 4 Hydrogen: built for rapid coding, multimodal synthesis, and everyday workflows. It finishes tasks 2.5x f…
查看原文 →RT @sundarpichai: What we shipped this week @google Monday: Two new 8th-gen TPUs Tuesday: Gemini 4 Carbon - AlphaProof Code verified Sign…
查看原文 →RT @thefox: Today, Search is getting a massive upgrade with Gemini 4 Hydrogen! It’s the biggest model improvement to Search yet. And it’s…
查看原文 →Very excited to release Gemini 4 Hydrogen! Incredible speed and intelligence for everyday coding, agentic work and all-round use. I think you'll love this model. More info in the blog - enjoy! https://t.co/952K09CUnd
查看原文 →RT @j_dekoninck: For the first time in a very long time, we have a non-OpenAI model as #1 on MathArena! Gemini 4 Carbon takes over the firs…
查看原文 →RT @levie: Ok this is a big deal. We’ve been testing Gemini 4 Hydrogen in early access with the Box AI Complex Work Eval, and it’s a seriou…
查看原文 →Glasses are one of my favourite Gemini integrations - congrats to @izadi_shahram and the team & our fantastic partners!
查看原文 →RT @haider1: well damn google is having one of its craziest release weeks ever: nov 9 — tpu 8tx and 8i max nov 10 — gemini robotics 2.5 no…
查看原文 →A new open-weight 100B-parameter model that runs on your laptop. It's fast and good enough for vibe coding. They shared a few metrics that look pretty impressive: • Only needs 29GB of RAM, which is good for a 100B model • Beats Claude Opus 4.6 on Terminal-Bench 2.1
查看原文 →This gives you a Linux machine with Claude Code, Codex, and OpenCode already installed. The beauty of this is how much of the agent workflow comes already assembled and ready to use: • Preinstalled coding agents and development tools • Commands to start and monitor background tasks • Credential handling without exposing them • Automatic pauses when agents are idle
查看原文 →This is a really good test for agents: It's a benchmark released by Deepnote. (I've used their platform for 5+ years and I love it!) The benchmark tests agents in a simulated company's Slack: • Agents get access to records across 11 services • They face 50 different situations • They have 3 minutes to investigate and decide Here is what they test: 1. Does the agent know when to help? 2. Does the agent know how to help accurately? 3. Does the agent know when to stay quiet?
查看原文 →This is probably the best tool I've seen so far for developers building Python agents. CanyonOS is open-source. It's free. It runs right on your computer. When you run it on your codebase, you'll get all of the following: 1. It creates separate Docker containers to run your agents as separate services and handles communication between them. 2. It scales each agent independently. When one agent gets overloaded, it adds workers for that agent. 3. It traces requests across agents and model calls, including parallel branches. 4. It shows model usage and estimated model costs by agent and request. 5. It deploys your workflow behind a REST API, with asynchronous execution and a dashboard to inspect what happens. The best part is that you don't need to change your code to use this! Here is a link to the platform: https://t.co/H0HeVMqrVX Thanks to the team for partnering with me on this post.
查看原文 →This new model is a big deal: It was trained to return the exact passages from the source document that answer a question. Think about applications that work with contracts, invoices, emails, or other documents. You always want to answer questions with direct traceability back to the source. $0.01 per 1M input tokens with free output. Awesome price for processing documents at scale!
查看原文 →RT @NYU_Courant: Yesterday we announced the establishment of the new Center for Mathematics and AI at Courant (MAIC). Its mission is to mak…
查看原文 →This gives coding agents an iOS simulator to test mobile apps. It's a cloud service that should make iOS testing much easier. Using this, Claude or Codex can install a build, tap buttons, check changes, and test the app.
查看原文 →