OpenAI launched GPT-5.4 — the first mainstream model with native PC control. The AI analyzes screenshots, clicks the mouse, types text, and executes commands without additional plugins. Context window expanded to 1.05M tokens (+160% vs GPT-5.2). GPT-5.4 Mini/Nano versions deliver <200ms response for real-time use. Available now in ChatGPT Pro ($200/month) and API. 🎯 PC Control: How It Works OSWorld-Verified benchmark: ✅ 75% success rate (beats average human) ✅ Opens browser → scrapes data → Excel → email ✅ Reads interface from screenshots, clicks buttons ✅ Shows action plan before execution
Technical specifications:
| Model | Context | Latency | Speed | API Price |
|---|---|---|---|---|
| GPT-5.4 | 1.05M | 450ms | 110 tok/s | $15/1M |
| GPT-5.4 Mini | 128K | 180ms | 185 tok/s | $3/1M |
| GPT-5.4 Nano | 64K | 45ms | 210 tok/s | $0.20/1M |
💻 PC Control Process
- Screenshot analysis → interface recognition
- GPT-5.4 creates action plan (shows to user)
- Click/type → verifies result on next screenshot
- User correction commands
- Safety: user policies by risk levels
Real use cases:
• Fill CRM → send 50 invoices to clients
• Scrape prices from 10 sites → Excel report
• 1C/QuickBooks: import → inventory reconciliation
🛡️ Pentagon Contract: $200M
OpenAI secured U.S. DoD contract:
Amount: $200M (June 2025)
Timeline: through July 2026
Purpose: AI prototypes for national security
Platform: OpenAI for Government
Military applications:
• Real-time satellite imagery analysis
• Logistics reporting automation
• Computer use for military interfaces
🔧 GPT-5.4 Technical Breakthroughs
1. Native computer use
PC control is built-in, not a plugin.
2. Multimodality
Screenshots + images + video + text.
3. Tool search
Auto-selects appropriate tools.
4. XHigh reasoning
Extreme reasoning for complex tasks.
💰 API Pricing
ChatGPT Pro: $200/month → GPT-5.4 + computer use
GPT-5.4 API: $15/1M input, $45/1M output
GPT-5.4 Nano: $0.20/1M — cheaper than GPT-4o mini
Rate limits (ChatGPT Pro):
GPT-5.4: 12K requests/day
GPT-5.4 Mini: 120K/day
GPT-5.4 Nano: 1.2M/day
🎯 Business Applications
Workflow automation
✅ CRM/ERP: form filling, reports
✅ Web scraping → Excel → distribution
✅ UI/UX testing from screenshots
✅ 1C/QuickBooks: data import/reconciliation
Developers:
✅ Playwright + GPT-5.4 = end-to-end tests
✅ Screenshot → code → deploy
✅ Auto-documentation from interfaces
⚔️ GPT-5.4 vs Competitors
| Model | Computer Use | Context | Latency | Price/1M |
|---|---|---|---|---|
| GPT-5.4 | ✅ Native | 1.05M | 450ms | $15 |
| Claude 3.7 Sonnet | Experimental | 200K | 850ms | $20 |
| Gemini 2.1 Pro | Via API | 1M | 650ms | $12 |
| Llama 4 | Plugins | 128K | 1200ms | Open |
🚀 OpenAI 2026 Roadmap
June: GPT-5.5 (2M context, 300ms)
September: Enterprise computer use
December: GPT-5.6 Nano (15ms latency)
Pentagon: prototypes by July 2026 → full deployment.
Publication author
