Model icon
gemini-4-argon
Mô hình:
Truy cập API Gemini 4 Argon của Google DeepMind trên PoYo qua OpenAI Chat Completions hoặc định dạng Gemini Native. Ngữ cảnh & đầu ra hợp nhất 1M, #1 trên Vals Index & DeepSWE v1.1 (77,9%), giá $1.60 đầu vào và $8.00 đầu ra mỗi 1M token (rẻ hơn 20% so với giá gốc).
Chat

Khu chơi này chỉ dành cho mục đích demo. Dữ liệu chỉ có giá trị trong cửa sổ hiện tại và sẽ được xóa khi cập nhật.

Nhập tin nhắn...

Thiết lập
1.0
1.0
Chi tiết về giá

Giá cả minh bạch mà không có phí ẩn. Chọn tiền đi.

Googlegemini-4-argon
Input
Giá PoYo
$1.60Các mã thông báo /1M
320 tín dụng
Giá chính thức
$2.00Các mã thông báo /1M
Official
Anh cứu
20%
Googlegemini-4-argon
Output
Giá PoYo
$8.00Các mã thông báo /1M
1600 tín dụng
Giá chính thức
$10.00Các mã thông báo /1M
Official
Anh cứu
20%
Googlegemini-4-argon
Cache Read
Giá PoYo
$0.160Các mã thông báo /1M
32 tín dụng
Giá chính thức
$0.200Các mã thông báo /1M
Official
Anh cứu
20%

* Các khoản phí thực tế được dựa trên kết quả cuối cùng.

GEMINI 4 ARGON · ĐÃ CÓ MẶT TRÊN POYO

Mô hình suy luận tiên phong của Google DeepMind với cửa sổ 1M token đầu vào-đầu ra hợp nhất cho lập trình, công việc doanh nghiệp và phòng thủ mạng

API Gemini 4 Argon cho Kỹ thuật Phần mềm Dài hạn, Quy trình Doanh nghiệp & Phòng thủ Mạng

Gemini 4 Argon là mô hình suy luận tiên phong sparse Mixture-of-Experts (MoE) thế hệ mới của Google DeepMind. Được xây dựng để duy trì suy luận sâu qua các tác vụ phức tạp, Argon mở rộng giới hạn token đầu ra lên mức dẫn đầu ngành là 1.000.000 token đầu vào-đầu ra hợp nhất (tăng từ 64K), cho phép tái cấu trúc toàn bộ kho mã nguồn, tổng hợp tài chính - pháp lý chuyên sâu và tự động vá lỗ hổng bảo mật chỉ trong một lượt.

Gọi gemini-4-argon trên PoYo qua POST /v1/chat/completions tương thích OpenAI hoặc POST /v1beta/models/gemini-4-argon:generateContent (và :streamGenerateContent). PoYo tính phí $1.60 / 1M token đầu vào (320 credit), $8.00 / 1M token đầu ra (1.600 credit) và $0.16 / 1M token đọc cache (32 credit)—thấp hơn 20% so với mức giá tiêu chuẩn của Google ($2.00 / $10.00 / $0.20).

Tính năng Nổi bật của API Gemini 4 Argon

Được chứng minh qua các bài đánh giá chính thức của Google DeepMind trong kỹ thuật phần mềm, phân tích doanh nghiệp, suy luận không gian 3D và an ninh mạng phòng thủ.

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

Bạn có thể xây dựng gì với API Gemini 4 Argon?

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.

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

Kết quả benchmark công bố tháng 10/2026 từ Model Card của Google DeepMind Gemini 4 Argon cùng bảng giá token chính thức và trên PoYo API.

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).

Cách gọi API Gemini 4 Argon trên PoYo

1. Tạo API Key: Đăng nhập vào bảng điều khiển PoYo và tạo khóa API.

2. Chọn giao thức Endpoint: Sử dụng POST https://api.poyo.ai/v1/chat/completions cho OpenAI SDK hoặc POST https://api.poyo.ai/v1beta/models/gemini-4-argon:generateContent (và :streamGenerateContent) cho định dạng Gemini gốc.

3. Đặt Model ID: Truyền "model": "gemini-4-argon" với hỗ trợ đầy đủ streaming (SSE).

Câu hỏi Thường gặp về API Gemini 4 Argon

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.

Tại sao nên tích hợp Gemini 4 Argon qua PoYo API?

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.