Google DeepMind · Dòng mô hình

Gemini

Google DeepMind’s Gemini models. Gemini 4 Carbon (November 2026) is the most capable model, for deep debugging, architecture, proofs and security audits; Gemini 4 Argon (September 2026) is Google’s recommended default for everyday coding and knowledge work. Both share a 1M-token window and the same per-token pricing, available through the Gemini API, Google AI Studio and Gemini Enterprise.

Chỉ API Văn bản / Đa phương thức Hiện tại: Gemini 4 Carbon

Bản phát hành hiện tại

Gemini 4 Carbon

Ổn định

Gemini 4 Carbon is Google’s strongest Gemini model, aimed at long-horizon agent work, hard software engineering, scientific reasoning and security tasks. It spends more tokens reasoning and verifying than Gemini 4 Argon, which Google still recommends for faster, lower-cost everyday coding and knowledge work.

Đã phát hànhtháng 11 10, 2026
Cửa sổ ngữ cảnh1M token
Đầu ra tối đa65,5K token
Giấy phépProprietary (Gemini API and Google apps)
Trọng sốChưa xuất bản

Thông số đã xác minh

Input price / 输入价格 USD 2 per million tokens until November 30, 2026; USD 4 from December 1 / 2026 年 11 月 30 日前每百万 token 2 美元,12 月 1 日起 4 美元 Nguồn →
Output price / 输出价格 USD 10 per million tokens (thinking included) until November 30, 2026; USD 20 from December 1 / 每百万 token 10 美元(含思考 token),12 月 1 日起 20 美元 Nguồn →
Cached input / 缓存输入 USD 0.10 per million tokens, USD 0.20 from December 1; implicit caching only / 每百万 token 0.1 美元,12 月 1 日起 0.2 美元;仅隐式缓存 Nguồn →
Free tier / 免费层 Not available on the Gemini API free tier / Gemini API 免费层不可用 Nguồn →
Token window / Token 窗口 1,048,576 tokens shared by input, thinking and output / 输入、思考和输出共用 1,048,576 token Nguồn →
Output limit / 输出上限 65,536 tokens per synchronous response; background mode can run to the remaining window / 同步请求最多 65,536 token;后台模式可用到窗口剩余长度 Nguồn →
Thinking levels / 思考档位 low, medium (default), high; temperature, top_p and top_k are ignored / low、medium(默认)、high;忽略 temperature、top_p、top_k Nguồn →
Where to use / 使用入口 Gemini API (Google AI Studio, Gemini Enterprise), Android Studio; Google AI Ultra in Antigravity, the Gemini app and AI Mode in Search / Gemini API(Google AI Studio、Gemini Enterprise)、Android Studio;AI Ultra 用户可在 Antigravity、Gemini 应用和搜索 AI 模式中使用 Nguồn →
Special access / 专项开放 Six months of free Antigravity access for approved scientists; Carbon for cyber through the Fairwind Program / 科研人员可申请半年免费 Antigravity 使用;安全团队可通过 Fairwind 计划申请 Carbon for cyber Nguồn →

Dòng thời gian phát hành

Gemini 4 Carbon
Ổn định · Đã phát hành thg 11 10, 2026

Gemini 4 Carbon is Google’s strongest Gemini model, aimed at long-horizon agent work, hard software engineering, scientific reasoning and security tasks. It spends more tokens reasoning and verifying than Gemini 4 Argon, which Google still recommends for faster, lower-cost everyday coding and knowledge work.

1M ngữ cảnh Chỉ API

Thông báo chính thức →

Gemini 4 Argon
Bản xem trước · Đã thông báo thg 9 30, 2026

Announced September 30, 2026. Limited rollout to selected trusted defenders through Fairwind and early testers; broader access is planned to start with paid API and Google AI Ultra users. Google states up to 1 million output tokens. A public API model ID, full rollout date and finalized API context specification are not confirmed in the reviewed sources.

Chỉ API

Thông báo chính thức →

Introducing Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber
Ổn định · Đã thông báo thg 7 21, 2026

We’re introducing new Gemini models, including Gemini 3.6 Flash, 3.5 Flash-Lite and 3.5 Flash Cyber.

Chỉ API

Thông báo chính thức →

Introducing Gemini 3.5 Flash Cyber
Ổn định · Đã thông báo thg 7 17, 2026

Google introduces Gemini 3.5 Flash Cyber, a lightweight cybersecurity model to find and patch vulnerabilities.

Chỉ API

Thông báo chính thức →

Start building with Nano Banana 2 Lite and Gemini Omni Flash
Ổn định · Đã thông báo thg 6 30, 2026
Chỉ API

Thông báo chính thức →

Introducing Gemini Omni
Ổn định · Đã thông báo thg 5 17, 2026
Chỉ API

Thông báo chính thức →

Gemini 3.5: frontier intelligence with action
Ổn định · Đã thông báo thg 5 15, 2026

Gemini 3.5 is built to help you execute complex, agentic workflows.

Chỉ API

Thông báo chính thức →

Gemini 3.1 Flash TTS: the next generation of expressive AI speech
Ổn định · Đã thông báo thg 4 15, 2026

Our newest audio model introduces granular audio tags that give you precise control to direct AI speech for expressive audio generation.

Chỉ API

Thông báo chính thức →

Gemini Robotics-ER 1.6: Powering real-world robotics tasks through enhanced embodied reasoning
Ổn định · Đã thông báo thg 4 13, 2026

Gemini Robotics ER 1.6: Enhancing spatial reasoning and multi-view understanding for autonomous robotics.

Chỉ API

Thông báo chính thức →

Gemini 3.1 Flash Live: Making audio AI more natural and reliable
Ổn định · Đã thông báo thg 3 26, 2026

Our latest voice model has improved precision and lower latency to make voice interactions more fluid, natural and precise.

Chỉ API

Thông báo chính thức →

Lyria 3 Pro: Create longer tracks in more
Ổn định · Đã thông báo thg 3 25, 2026

Introducing Lyria 3 Pro, which unlocks longer tracks with structural awareness. We’re also bringing Lyria to more Google products and surfaces.

Chỉ API

Thông báo chính thức →

Measuring progress toward AGI: A cognitive framework
Ổn định · Đã thông báo thg 3 17, 2026

We’re introducing a framework to measure progress toward AGI, and launching a Kaggle hackathon to build the relevant evaluations.

Chỉ API

Thông báo chính thức →

Gemini 3.1 Flash-Lite: Built for intelligence at scale
Ổn định · Đã thông báo thg 3 03, 2026

Gemini 3.1 Flash-Lite is our fastest and most cost-efficient Gemini 3 series model yet.

Chỉ API

Thông báo chính thức →

Gemini 3.1 Pro: A smarter model for your most complex tasks
Ổn định · Đã thông báo thg 2 19, 2026

3.1 Pro is designed for tasks where a simple answer isn’t enough.

Chỉ API

Thông báo chính thức →

Dự án

google-deepmind/mujoco Chính thức

A multijoint contact dynamics engine maintained by Google DeepMind, providing model description, physics stepping, an interactive viewer, and a Python interface. It is suitable for developing robotics, control, and reinforcement learning environments.

★ 15,1K · ⑂ 1,7K

OpenSpiel is a collection of environments and algorithms for research in general reinforcement learning and search/planning in games.

★ 5,4K · ⑂ 1,2K

Bài viết liên quan

Gemini 4 Carbon opens to paying developers at $2/$10 per million tokens until November 30, then doubles thg 11 11, 2026 Gemini 4 Carbon 开放付费 API:11 月 30 日前每百万 token 2/10 美元,之后翻倍 thg 11 11, 2026 AI Ultra 的 1.2 亿 Opus 5.5 额度值 50 美元?先把口径算清楚 thg 10 04, 2026 Gemini 4 Argon:我更想知道,它能替我少返几次工 thg 10 03, 2026 Google expands AI & Economy ATLAS with interactive data and research on scientists’ AI use thg 9 29, 2026 OpenAI contractors review real ChatGPT chats, investigation finds thg 9 15, 2026 Gemini 3.8 Flash Takes the Price War Down to Token Consumption thg 9 04, 2026 Google’s Latest AI Announcements from August 2026 thg 9 02, 2026 Get closer to the game with Gemini and Pixel thg 8 18, 2026 AlphaEarth Foundations: A New AI Model for Detailed Global Mapping thg 8 18, 2026 Bring your spreadsheet data to life with Sheets canvas thg 8 15, 2026 Google DeepMind Expands in Singapore to Advance AI in Asia-Pacific thg 8 15, 2026