GPT Image 2.5 API が登場。画像生成・編集を今すぐ体験 →
PoYoPoYo
Model icon
gemini-4-argon
Chat
Google の他のモデルを見る
モデル:
Google DeepMind の最新フラッグシップ Gemini 4 Argon API を PoYo で利用可能。OpenAI 互換および Gemini ネイティブ形式に対応。1M 統合入出力コンテキスト、Vals Index と DeepSWE v1.1 (77.9%) で1位、入力 $1.60・出力 $8.00 / 100万トークン（公式標準価格の20%オフ）。
APIキー取得
APIドキュメント
Chat

この Playground はデモ用です。データは現在のウィンドウ内でのみ有効で、更新するとクリアされます。

メッセージを入力...

設定
1.0
1.0
価格詳細

透明な価格設定。隠れた費用はありません。使った分だけ支払えます。

Googlegemini-4-argon
Input
PoYo価格
$1.60/1M tokens
320 クレジット
公式価格
$2.00/1M tokens
Official
節約率
20%
Googlegemini-4-argon
Output
PoYo価格
$8.00/1M tokens
1600 クレジット
公式価格
$10.00/1M tokens
Official
節約率
20%
Googlegemini-4-argon
Cache Read
PoYo価格
$0.160/1M tokens
32 クレジット
公式価格
$0.200/1M tokens
Official
節約率
20%

* 実際の料金は最終出力に基づきます。

おすすめモデル
GEMINI 4 ARGON · POYO で提供開始

100万トークンの統合入出力ウィンドウを備え、長期ソフトウェア開発、専門知識業務、サイバー防衛を革新する Google DeepMind の最先端推論モデル

長期コーディング・エンタープライズ分析・サイバー防衛のための Gemini 4 Argon API

Gemini 4 Argon は、Google DeepMind が発表した次世代のスパース Mixture-of-Experts（MoE）フロンティア推論モデルです。複雑で長期的なワークフローでの深い推論を実現するため、最大出力上限を従来の 64K から業界最高水準の 1,000,000（1M）統合入出力トークンへと大幅に拡張し、リポジトリ全体のリファクタリングや大規模な金融・法務分析、脆弱性の自律修正を1回の推論で完結できます。

PoYo ではモデル ID gemini-4-argon を使用し、OpenAI 互換の POST /v1/chat/completions または Gemini ネイティブの POST /v1beta/models/gemini-4-argon:generateContent（および :streamGenerateContent）から呼び出せます。料金は入力 $1.60 / 1M トークン（320 クレジット）、出力 $8.00 / 1M トークン（1,600 クレジット）、キャッシュ読込 $0.16 / 1M トークン（32 クレジット）で、Google 公式標準価格（$2.00 / $10.00 / $0.20）より 20% お得です。

PoYo API キーを取得

Gemini 4 Argon API の主な特徴

ソフトウェアエンジニアリング、金融・法務の知識業務、3D・長尺動画のマルチモーダル理解、防御的サイバーセキュリティにおける Google DeepMind 公式評価と実績。

01

State-of-the-Art Long-Horizon Software Engineering (77.9% DeepSWE v1.1)

Gemini 4 Argon sets a new state of the art on DeepSWE v1.1 (77.9%) and Vibe Code Bench (91.9%), outperforming Claude Opus 5.5 (74.2%) and GPT-6 Astra (74.1%). With a 1M token output limit, Argon can untangle multi-million-line legacy codebases, migrate C/C++ libraries to safe Rust (achieving a 2.7x speedup in Google's libgav1 video decoder), and optimize vLLM GPU kernels end-to-end without mid-response truncation.

  • 77.9% on DeepSWE v1.1 and 91.9% on Vibe Code Bench
  • Up to 1,000,000 unified input and output tokens per trajectory
  • Multi-file architecture refactoring and C++ to Rust migrations
Official Google DeepMind DeepSWE v1.1 benchmark chart showing Gemini 4 Argon at 77.9% vs Claude Opus 5.5 and GPT-6 Astra

02

#1 on Economic Knowledge Work & Enterprise Automation

Argon ranks #1 on the Vals Index (68.9%)—weighted across finance, legal, tax, and coding by U.S. GDP contribution—and leads Vals Finance Agent v2 (65.4%), Harvey's Legal Agent Benchmark (19.6% vs 5.4% for Astra), and Zapier's AutomationBench (51.3%). Its GraphWalks 256K–1M long-context recall reaches 84.2% F1, over 17 percentage points ahead of competing frontier models.

  • 68.9% on Vals Index and 65.4% on Vals Finance Agent v2
  • 19.6% on Harvey's Legal Agent Benchmark (3x higher than peers)
  • 84.2% F1 on GraphWalks 256K–1M long-context reasoning
Official Google DeepMind Vals Index benchmark comparison showing Gemini 4 Argon leading at 68.9%

03

Multimodal Video, 3D Spatial & Scientific Synthesis

Natively trained across text, images, audio, video, and PDFs, Gemini 4 Argon achieves 91.7% on LVBench (long video understanding), 88.8% on LABBench 2, and 87% on Embodied Reasoning (ERQA). In official DeepMind demonstrations, Argon synthesizes 50 GB of Airborne LiDAR, GIS vectors, and archival CAD drawings into interactive 3D reconstructions, and drives multi-agent visual critique loops in Blender and Three.js.

  • 91.7% on LVBench long-video comprehension and 71.6% on Chartography
  • 88.8% on LABBench 2 and 76.0% on RiemannBench mathematical reasoning
  • Autonomous 3D scene generation and multimodal scientific synthesis

04

Defensive Cybersecurity & Prompt Injection Resilience

Built for cyber defenders in Google's Fairwind Program and Wiz's Scan for Good initiative, Argon ties for #1 on CWE-bench v1 (68.0%) for autonomous vulnerability remediation and discovers complex multi-language exposures across 20 programming languages. Adversarial training reduces indirect prompt injection success to just 0.7% across 15 adaptive attempts on the Gray Swan IPI benchmark.

  • 68.0% on CWE-bench v1 for automated security patching
  • 0.7% attack success rate on Gray Swan Indirect Prompt Injection (IPI)
  • Autonomous source-code auditing and proof-of-concept validation
Official Google DeepMind CWE-bench v1 cybersecurity vulnerability remediation benchmark chart for Gemini 4 Argon

Gemini 4 Argon API の活用ユースケース

01

Full-Repository Refactoring & Language Migration

Leverage the 1M unified input-output token window to refactor monolithic modules, migrate C/C++ codebases to memory-safe Rust, and output complete multi-file patches in one trajectory.

02

Autonomous Enterprise & Operations Agents

Orchestrate end-to-end business workflows across APIs, databases, and document repositories with #1 AutomationBench (51.3%) reliability and transparent chain-of-thought monitoring.

03

Financial Modeling & Legal Due Diligence

Connect evidence across hundreds of SEC filings, contracts, and case precedents with leading accuracy on Vals Finance Agent v2 (65.4%) and Harvey's Legal Agent Benchmark (19.6%).

04

Automated Security Auditing & Patching

Scan multi-language repositories to surface subtle logic flaws, validate exploitability in sandboxed environments, and generate verified CWE remediation patches.

05

Long-Video & Multimodal Document Intelligence

Analyze hours-long video recordings, technical charts, audio streams, and dense PDFs with 91.7% LVBench accuracy and 84.2% recall across 256K–1M token contexts.

06

Scientific Computing & 3D Simulation

Accelerate quantum circuit optimization, molecular conformer search, formal proof verification in Lean, and browser-ready Three.js / WebGL scientific visualizations.

Gemini 4 Argon vs GPT-6 Astra vs Claude Opus 5.5 vs Claude Fable 5.1 比較

Google DeepMind の Gemini 4 Argon Model Card（2026年10月）の公式ベンチマーク結果と、公式および PoYo API のトークン料金比較。

Benchmark / MetricGemini 4 ArgonGPT-6 AstraClaude Opus 5.5Claude Fable 5.1
Vals Index (Knowledge Work)68.9% (#1)63.1%67.0%65.8%
AutomationBench (End-to-End Workflows)51.3% (#1)41.4%42.5%31.4%
Vals Finance Agent v265.4% (#1)53.5%58.6%58.9%
Harvey's Legal Agent Benchmark19.6% (#1)5.4%3.8%6.7%
DeepSWE v1.1 (Agentic Coding)77.9% (#1)74.1%74.2%67.4%
Vibe Code Bench91.9% (#1)89.6%90.3%90.3%
Terminal-Bench 4.057.4%58.2%66.4%57.9%
LABBench 2 / RiemannBench (Science & Math)88.8% / 76.0%85.4% / 72.0%73.1% / 69.6%68.6% / 65.6%
GraphWalks 256K–1M (Long Context F1)84.2% (#1)71.8%66.8%65.0%
LVBench (Long Video) / Chartography91.7% / 71.6%87.5% / 71.0%83.7% / 66.3%79.7% / 46.2%
CWE-bench v1 (Cybersecurity)68.0% (Tied #1)68.0%67.0%58.0%
Context Window / Max Output1M / 1M unified1.05M / 128K1M / 128K1M / 128K
Official Standard Price (Input / Output per 1M)$2.00 / $10.00$10.00 / $50.00$4.00 / $20.00$10.00 / $50.00
PoYo API Price (Input / Output / Cache Read per 1M)$1.60 / $8.00 / $0.16$8.00 / $40.00 / $0.80$3.20 / $16.00 / $0.16$8.00 / $40.00 / $0.20

Source: Google DeepMind Gemini 4 Argon Model Card (October 2026). PoYo pricing reflects a 20% discount off official standard rates (200 credits = $1.00 USD).

PoYo で Gemini 4 Argon API を利用する方法

1. API キーの作成：PoYo ダッシュボードにログインして API キーを発行します。

2. エンドポイント形式の選択：OpenAI SDK 互換の POST https://api.poyo.ai/v1/chat/completions、または Gemini ネイティブの POST https://api.poyo.ai/v1beta/models/gemini-4-argon:generateContent（ストリーミングは :streamGenerateContent）を選択します。

3. モデル ID の指定："model": "gemini-4-argon" を指定してリクエストを送信します。

Gemini 4 Argon API に関するよくある質問

What is Gemini 4 Argon?

Gemini 4 Argon is Google DeepMind's flagship sparse Mixture-of-Experts (MoE) frontier reasoning model announced in late September 2026 and released globally in October 2026. It is engineered for long-horizon software engineering, high-stakes enterprise knowledge work (finance, legal, tax), scientific discovery, and defensive cybersecurity.

How much does the Gemini 4 Argon API cost on PoYo vs Google Official?

Google's official standard introductory pricing is $2.00 per 1M input tokens, $10.00 per 1M output tokens, and $0.20 per 1M cached input tokens. On PoYo, you get a flat 20% discount: $1.60 / 1M input tokens (320 credits), $8.00 / 1M output tokens (1,600 credits), and $0.16 / 1M cache read tokens (32 credits), where 200 credits = $1.00 USD.

Which API endpoints does PoYo support for Gemini 4 Argon?

PoYo supports both the OpenAI-compatible Chat Completions endpoint (POST https://api.poyo.ai/v1/chat/completions) and the Google Gemini Native endpoints (POST https://api.poyo.ai/v1beta/models/gemini-4-argon:generateContent and :streamGenerateContent) using the model ID gemini-4-argon.

What is the context window and maximum output token limit of Gemini 4 Argon?

Gemini 4 Argon features a 1,000,000-token unified input-output context window and expands maximum output capacity to 1,000,000 tokens (up from 64K in previous Gemini models). This allows the model to sustain 100K+ token reasoning chains and generate full-repository code or exhaustive reports in a single trajectory.

How does Gemini 4 Argon compare to Claude Opus 5.5 and GPT-6 Astra?

On Google DeepMind's published evaluations, Gemini 4 Argon leads on DeepSWE v1.1 (77.9% vs 74.2% for Opus 5.5 and 74.1% for Astra), Vals Index (68.9% vs 67.0% and 63.1%), AutomationBench (51.3%), GraphWalks 256K–1M (84.2%), and LVBench (91.7%), while tying Astra on CWE-bench v1 (68.0%). Claude Opus 5.5 leads on Terminal-Bench 4.0 (66.4% vs 57.4%), and GPT-6 Astra leads on FrontierSWE v2 (65.5%) and Terminal-Bench Science 0.1 (68.1%).

What input modalities and knowledge cutoff does Gemini 4 Argon have?

According to the Gemini 4 Argon Model Card, the model natively accepts text, images, audio, video, and PDF documents as input and produces text output. Its knowledge cutoff is March 2026.

PoYo で Gemini 4 Argon API を統合する理由

01

20% Lower Token Costs

Pay $1.60 input, $8.00 output, and $0.16 cache read per 1M tokens—saving 20% compared to official rates with flexible pay-as-you-go credits.

02

Dual OpenAI & Gemini Native Formats

Switch seamlessly between /v1/chat/completions for existing OpenAI SDKs and /v1beta/models/gemini-4-argon:generateContent for native Gemini features.

03

Benchmark Side-by-Side with Opus & Astra

Use one PoYo API key to route workloads across Gemini 4 Argon, Claude Opus 5.5, Claude Sonnet 5.5, GPT-6 Astra, and GPT-6.1 Sol.

04

Instant Global Access, No Tier Gating

Start calling gemini-4-argon immediately in production or testing in the web Playground without enterprise waitlists or regional quota bottlenecks.