Gemini 4 Argon
Google DeepMind's new flagship for code, agents and complex documents
無料で利用できます
- コンテキスト
- 1,048,576 tokens
- 入力
- テキスト、画像、PDF
- タイプ
- 推論対応
- リリース
- October 2026
- 開発元
- Google DeepMind
- 利用
- ChatPlus PRO
Gemini 4 Argon is the first model of the fourth Gemini generation, opened to everyone on October 13, 2026. Google built it for long, demanding work: refactoring large codebases, multi-step agentic tasks, financial and legal analysis. It understands text, images and PDFs, with a context of a million tokens. Available on ChatPlus on the PRO plan.
Argon is the first Gemini 4 model. Google unveiled it on September 30 but at first gave it only to cybersecurity specialists and trusted testers. For two weeks the model was tested on real work and its safeguards were tuned, and on October 13 it opened to everyone: in the Gemini app, the API and the Antigravity development environment.
Inside Google, Argon is already rewriting large projects. The Chrome team used it to untangle a six-million-line folder of browser features: a refactor across hundreds of files took weeks instead of months. The model was trained to say so honestly when something did not work, and to stop and ask for confirmation before an irreversible step.
It is not only about code. In Google's tests Argon beats its rivals at financial analysis, legal work and understanding long videos, and it writes more directly than earlier Gemini models: less hedging, fewer repeats and clichés. It still trails in terminal tasks, where Claude Opus 5.5 leads for now.
On ChatPlus, Gemini 4 Argon is available on the PRO plan. Use it for large documents, code and analysis — and compare it with GPT, Claude and Grok in the same window.
何が得意か
プログラミング
77.9% on DeepSWE v1.1, the best score at launch: ahead of GPT-6 Astra (74.1%) and Claude Opus 5.5 (74.2%).
Document work
First place on the Vals Index at 68.9%. The test is built from real tasks in finance, law, tax and code.
エージェントタスク
51.3% on AutomationBench against 42.5% for Opus 5.5: it carries multi-step business workflows through to the end.
長いコンテキスト
84.2% on GraphWalks in the 256K–1M range. GPT-6 Astra scores 71.8% on the same test.
Gemini 4 Argonも他のすべてのモデルも——1つのプラン、1つの画面で。
特に強い領域
Geminiと他社モデルを比較
Gemini 4 Argonと他社ラボの主力モデルを横並びで比較できます。すべてChatPlusのサブスクリプションに含まれています。
| Claude | DeepSeek | GPT | Gemini | Grok | GLM | Kimi | Qwen | Muse | GLM 5.3 Flash | |
|---|---|---|---|---|---|---|---|---|---|---|
| 開発元 | Anthropic | DeepSeek | OpenAI | xAI | Z.ai | Moonshot | Alibaba | Meta | Z.ai | |
| モデル | Sonnet 4.6 | DeepSeek V4.1 Reasoner | GPT-6 Astra | Gemini 4 Argon | Grok 4 | GLM 5 | Kimi K3 | Qwen 3.8 Max | Muse Spark 1.3 | GLM 5.3 Flash |
| コンテキスト | 200K | 128K | 1M | 1M | 256K | 1M | 256K | 256K | 1M | 1M |
| 画像を理解 | — | — | ||||||||
| 最新のWebデータ | — | — | — | — | — | — | — | |||
| オープンソース | — | — | — | — | — | |||||
| 主な強み | テキストとコード | 推論 | コンピューター操作 | コード | 速報性 | コードとエージェント | 長いコンテキスト | 複雑なタスク | エージェントとコード | コードとエージェント |
| ChatPlusで |