OpenAI has released GPT-5.4, its most capable and efficient frontier model yet, available in ChatGPT (as GPT-5.4 Thinking), the API, and Codex. The AI major has also launched a high-performance variant, GPT-5.4 Pro, for complex tasks.
The upgrade integrates top advances in reasoning, coding, and agentic workflows, incorporating GPT-5.3-Codex's coding strengths while enhancing tool use, software environments, and professional tasks like spreadsheets, presentations, and documents. It delivers more accurate results with minimal back and forth, OpenAI said.
ChatGPT GPT-5.4: Key New Capabilities
As per OpenAI, key features and benefits of ChatGPT GPT-5.4 include:
- 1 million token context window (in API and Codex), expanding the context window and enabling the model to process, remember, and analyse longer conversations or documents and verification for agents.
- Native computer-use capabilities — the first in a general-purpose OpenAI model — allowing agents to operate computers via screenshots, mouse/keyboard actions, and libraries like Playwright. OpenAI highlighted that the model excels on benchmarks: 75% success rate on OSWorld-Verified versus GPT-5.2's 47.3% and surpassing human performance at 72.4%, 67.3% on WebArena-Verified, and 92.8% on Online-Mind2Web with screenshots.
- Tool search for efficient discovery and use across large tool ecosystems.
- GPT-5.4 Thinking mode provides its thinking plan upfront, allowing mid-response adjustments for better alignment of output with intention without extra iterations.
- Improved deep web research, context maintenance for longer thinking, and multi-source synthesis for specific queries.
- Improved token efficiency, using far fewer tokens than GPT-5.2 for faster and cheaper reasoning.
- “Most factual model yet,” says OpenAI, reducing hallucinations, with 33% fewer false claims and 18% fewer erroneous responses versus GPT-5.2.
- Improved professional performance, with OpenAI saying the model matches/exceeds humans in 83% of comparisons on GDPval (vs. 70.9% for GPT‑5.2), 87.3% on spreadsheet tasks (vs. 68.4% for GPT‑5.2), and is preferred for presentations 68% of the time.
- GPT-5.4 expands reliable agents, developer workflows, and high-quality outputs in coding, document handling, business analytics, finance, and more.
Also read: Iran Conflict Could Hit Samsung, Other South Korean Chipmakers: Report
Essential Business Intelligence, Sharp Market Insights, Practical Personal Finance Advice, Daily Fuel, Gold and Silver Prices and Latest Stories — On NDTV Profit.
