OpenAI Launches GPT-5.4: Native Computer Use, 1M Context Window, Enhanced Coding

OpenAI officially released GPT-5.4 on March 5, 2026, calling it the 'most capable and efficient frontier model for professional work.' The model ships in three variants: standard GPT-5.4, GPT-5.4 Thinking (with enhanced reasoning), and GPT-5.4 Pro (maximum performance), none available to free-tier users. GPT-5.4 consolidates the coding prowess of GPT-5.3-Codex while upgrading reasoning, tool use, and professional workflows.

Key breakthroughs include native computer-use capabilities, enabling the model to operate websites and software via code or keyboard/mouse commands, achieving 75.0% on OSWorld-Verified (surpassing the 72.4% human baseline); a 1.05M-token context window for handling long documents and complex multi-step tasks; and a new Tool Search feature that reduces API token usage by 47%. On the GDPval benchmark across 44 occupations, GPT-5.4 matches or exceeds industry professionals in 83.0% of comparisons.

Pricing stands at $2.50/$15.00 per million tokens (input/output) for the standard model and $30.00/$180.00 for Pro. The model is now available across ChatGPT, Codex, OpenAI's API, Microsoft 365 Copilot, and Copilot Studio.

OpenAI GPT-5.4 Deep Research Report: A Pivotal Step Toward Practical AI Agents

1. Background

On March 5, 2026, OpenAI officially released GPT-5.4, the fourth major model version in under five months following GPT-5.1 (November 2025), GPT-5.2 (December 2025), and GPT-5.3-Codex (February 2026). This aggressive release cadence reflects the intensity of the current AI race, with Anthropic's Claude Opus 4.6 and Google's Gemini 3 Pro competing fiercely in the same timeframe.

OpenAI positions GPT-5.4 as its 'most capable and efficient frontier model for professional work,' explicitly targeting professional users, developers, and enterprise applications. Notably, all three variants — standard GPT-5.4, GPT-5.4 Thinking, and GPT-5.4 Pro — are unavailable to free-tier users, signaling OpenAI's continued strategic shift toward the premium market.

2. Core Technical Breakthroughs

#### 2.1 Native Computer Use

GPT-5.4 is OpenAI's first universal model with native computer-use capabilities. The model can directly interact with software environments, applications, and websites by writing code or issuing keyboard and mouse commands to autonomously perform multi-step tasks across different platforms. On the OSWorld-Verified benchmark, GPT-5.4 achieved a 75.0% success rate — surpassing the human baseline of 72.4%.

This capability directly competes with Anthropic's Claude Computer Use feature, though OpenAI has chosen to deeply integrate it into the universal model rather than offering it as a separate capability.

#### 2.2 Million-Token Context Window

GPT-5.4 supports a context window of up to 1.05 million tokens. This enables the model to maintain coherence across extended reasoning chains and effectively handle document-heavy and analytical workflows. However, requests exceeding 272K input tokens are billed at double the rate.

#### 2.3 Tool Search

GPT-5.4 introduces Tool Search, which allows the model to search for specific tools on demand rather than requiring all tool definitions upfront. In tests with 250 tasks from the MCP Atlas benchmark, Tool Search reduced token usage by 47% while maintaining accuracy.

#### 2.4 Enhanced Reasoning — Thinking Mode

GPT-5.4 Thinking dedicates more computational resources to complex problem-solving. Its innovation lies in providing an upfront plan of its thinking process, allowing users to adjust the model's course mid-response.

3. Performance Benchmarks

| Metric | GPT-5.4 | GPT-5.2 | Improvement |

|--------|---------|---------|-------------|

| False claim probability | — | — | 33% reduction |

| Response error rate | — | — | 18% reduction |

| GDPval professional work | 83.0% | 70.9% | +12.1pp |

| OSWorld-Verified | 75.0% | — | Exceeds human baseline (72.4%) |

| SWE-Bench Pro coding | 57.7% | — | Lower latency |

| MMMU-Pro visual understanding | 81.2% | — | — |

| Spreadsheet tasks | 87.3% | — | — |

4. Pricing Strategy

|-------|---------------------|--------|----------------------|

| GPT-5.4 Standard | $2.50 | $0.25 | $15.00 |

| GPT-5.4 Pro | $30.00 | N/A | $180.00 |

| Claude Opus 4.6 | $5.00 | — | $25.00 |

| Gemini 3 Pro | $2.00 | — | $12.00 |

5. Industry Impact

Professional Work Automation: GDPval covers 44 occupations with 83% match/exceed rate, indicating deepening AI penetration in knowledge work.

Practical AI Agents: Native computer-use exceeding human baselines means AI agents have real-world deployment potential, driving transformation in the RPA industry.

Developer Ecosystem: Integration of Codex coding + Tool Search cost reduction + million-token context creates a compelling developer ecosystem.

6. Competitive Landscape

The current frontier model competition is three-way:

**OpenAI GPT-5.4:** Professional workflows, computer use, coding integration
**Anthropic Claude Opus 4.6:** Collaborative coding via Claude Code and Cowork
**Google Gemini 3 Pro:** Multimodal strength and price competitiveness

GPT-5.4 directly targets Anthropic's Claude series, particularly in coding and professional work. Microsoft's integration of GPT-5.4 Thinking into 365 Copilot further consolidates the OpenAI-Microsoft alliance in enterprise AI.

References

1. [Wikipedia - GPT-5.4](https://en.wikipedia.org/wiki/GPT-5.4)

2. [Mashable](https://mashable.com/article/gpt-5-4-release-improvements-changes)

3. [Tom's Guide](https://www.tomsguide.com/ai/gpt-5-4-is-here)

4. [TechCrunch](https://techcrunch.com/2026/03/05/openai-launches-gpt-5-4/)

5. [Ars Technica](https://arstechnica.com/ai/2026/03/openai-introduces-gpt-5-4/)