Overview
GPT-5.4 is OpenAI's most capable frontier model for professional work, released on March 5, 2026. It unifies the best of recent advances in reasoning, coding, and agentic workflows into a single model—incorporating the industry-leading coding capabilities of GPT-5.3-Codex while adding native computer-use and significantly improved tool orchestration. Available as GPT-5.4 Thinking in ChatGPT, and as gpt-5.4 in the API and Codex, this release is positioned as OpenAI's strongest mainline reasoning model across professional work, computer use, coding, and tool-heavy workflows, based on multiple official evaluations.
What's New
Professional Knowledge Work
GPT-5.4 sets a new standard for real-world professional tasks. On GDPval—which tests agents across 44 occupations spanning the top 9 industries contributing to U.S. GDP—GPT-5.4 matches or exceeds industry professionals in 83.0% of comparisons, up from 70.9% for GPT-5.2. Tasks include creating sales presentations, accounting spreadsheets, manufacturing diagrams, and short videos.
Factual accuracy also improves substantially: GPT-5.4 produces individual claims that are 33% less likely to be false and full responses that are 18% less likely to contain any errors compared to GPT-5.2. On spreadsheet modeling tasks representative of junior investment banking analyst work, GPT-5.4 scores 87.3% versus GPT-5.2's 68.4%.
Native Computer-Use Capabilities
GPT-5.4 is OpenAI's first general-purpose model with native, state-of-the-art computer-use capabilities. It achieves 75.0% on OSWorld-Verified—which tests navigating a desktop environment via screenshots and keyboard/mouse actions—surpassing human performance at 72.4% and far exceeding GPT-5.2's 47.3%.
Browser use also improves: on WebArena-Verified, GPT-5.4 reaches 67.3% using combined DOM- and screenshot-driven interaction. In customer-reported evals across ~30K HOA and property tax portals, GPT-5.4 achieved a 95% first-attempt success rate, completing sessions 3× faster while using 70% fewer tokens compared to prior computer-use models.
Developers can access computer-use capabilities via the updated computer tool in the API.
Steerability and Transparent Thinking
In ChatGPT, GPT-5.4 Thinking provides an upfront plan of its approach before working on complex queries. Users can adjust direction mid-response and steer the model toward their intended outcome without starting over. The model also handles longer workflows with stronger context awareness, maintaining coherence across extended conversations and multi-step tasks.
Tool Search for Large Tool Ecosystems
GPT-5.4 introduces tool search in the API, allowing the model to work efficiently with large collections of tools by looking up tool definitions on demand rather than loading all definitions upfront.
In testing across 250 tasks from Scale's MCP Atlas benchmark with all 36 MCP servers enabled, the tool-search configuration reduced total token usage by 47% while achieving identical accuracy. For MCP servers with tens of thousands of tokens of tool definitions, efficiency gains can be even larger.
Coding Capabilities
GPT-5.4 combines GPT-5.3-Codex's coding strengths with computer-use and professional knowledge, enabling longer-running tasks where the model can use tools, iterate, and push work further with less manual intervention. It matches or outperforms GPT-5.3-Codex on SWE-Bench Pro (57.7% vs 56.8%), while running at lower latency across all reasoning effort levels.
In Codex, /fast mode delivers up to 1.5× faster token velocity with GPT-5.4 for coding tasks, debugging, and iteration. Developers can also access this speed through the API with priority processing.
1M Token Context Window
GPT-5.4 supports up to 1M tokens of context in Codex (experimental), enabling agents to plan, execute, and verify tasks across long horizons. Requests exceeding the standard 272K window count against usage limits at 2× the normal rate.
Improved Web Search and Browsing
On BrowseComp—which measures how well AI agents persistently browse the web to find hard-to-locate information—GPT-5.4 jumps 17 percentage points over GPT-5.2 (82.7% vs 65.8%). GPT-5.4 Pro sets a new state of the art at 89.3%, making this version substantially stronger for research tasks requiring synthesis from many diverse sources.
Enhanced Visual Perception
GPT-5.4 improves visual understanding for high-resolution images with a new original image detail level supporting full-fidelity perception up to 10.24M total pixels or 6000-pixel maximum dimension. On MMMU-Pro (visual understanding and reasoning), GPT-5.4 achieves 81.2% without tool use, up from GPT-5.2's 79.5%.
Availability & Access
Launch Availability (Mar 5, 2026)
GPT-5.4 began rolling out on March 5, 2026. At launch, GPT-5.4 Thinking in ChatGPT became available to Plus, Team, and Pro users, replacing GPT-5.2 Thinking. GPT-5.2 Thinking remains available under Legacy Models for three months, with planned retirement on June 5, 2026. Enterprise and Edu plan admins could enable early access via admin settings. At launch, GPT-5.4 Pro rolled out to Pro and Enterprise plans; current official plan pages also list Business, and Help documentation includes Edu.
Current Plan Availability
In ChatGPT, GPT-5.4 Thinking is accessible across Free, Plus, Team, Pro, Business, Enterprise, and Edu plans (usage limits and feature access vary by tier). GPT-5.4 Pro remains restricted to higher-tier plans—refer to the official pricing page for the current access matrix.
In the API, gpt-5.4 is broadly available across Chat Completions, Responses, and Realtime endpoints, while gpt-5.4-pro is currently available in the Responses API only.
Pricing & Plans
GPT-5.4 follows OpenAI's existing subscription structure with API pricing updates to reflect improved capabilities:
- ChatGPT Free — $0/month, limited access to GPT-5.4 Thinking with usage limits.
- ChatGPT Go — $8/month, includes GPT-5.4 Thinking access (check official page for current limits).
- ChatGPT Plus — $20/month, expanded access to GPT-5.4 Thinking with higher usage limits.
- ChatGPT Pro — $200/month, access to GPT-5.4 Pro and GPT-5.4 Thinking with extended limits and priority access.
- ChatGPT Business — Team workspace with admin features; pricing varies by seat count.
- ChatGPT Enterprise/Edu — Custom pricing; early access via admin settings.
API pricing for GPT-5.4:
| Model | Input | Cached Input | Output |
|---|---|---|---|
| gpt-5.4 | $2.50/M tokens | $0.25/M tokens | $15/M tokens |
| gpt-5.4-pro | $30/M tokens | — | $180/M tokens |
Batch and Flex processing are available at half the standard rate. Priority processing is available at 2× the standard rate. Despite higher per-token pricing, GPT-5.4's greater token efficiency typically reduces total token usage—especially in tool-heavy workflows.
Best For
- Enterprise professionals creating spreadsheets, presentations, and documents who need AI that matches the accuracy of junior analysts and industry specialists.
- Developers building computer-use agents who require a model that can operate browsers, desktop software, and complex UI workflows with near-human success rates—enabled by GPT-5.4's native computer-use achieving 75.0% on OSWorld-Verified.
- Teams with large MCP or tool ecosystems who need efficient, accurate tool orchestration across dozens of connectors without prohibitive token overhead. For background on MCP vs A2A protocols, see our comparison guide.
- Research analysts and legal professionals tackling long-horizon document analysis, contract review, and multi-step information synthesis tasks.
- Software engineers using Codex who want frontier coding capabilities combined with computer-use for visual debugging and automated browser playtesting.
- Product and content teams seeking higher-quality presentations and documents with improved aesthetic output and more effective use of image generation.
FAQ
What is GPT-5.4 Thinking in ChatGPT?
GPT-5.4 Thinking is the ChatGPT-facing version of GPT-5.4. It shows an upfront plan of its reasoning approach for complex queries, allowing you to adjust direction mid-response. This is available to Plus, Team, and Pro users starting March 5, 2026, replacing GPT-5.2 Thinking.
How does GPT-5.4's computer use work for developers?
Developers access computer-use capabilities through the updated computer tool in the API. GPT-5.4 can interpret screenshots and issue coordinate-based mouse and keyboard commands. Behavior is steerable via developer messages, and custom confirmation policies can be configured for different risk levels. See OpenAI's documentation for recommended best practices.
Is GPT-5.4 worth the higher API cost compared to GPT-5.2?
For many workflows, yes—especially tool-heavy agentic tasks. Tool search reduces token usage by up to 47% on MCP workflows, and the model's improved accuracy means fewer correction loops. For straightforward tasks, GPT-5.2 remains available and may be more cost-effective.
When will GPT-5.2 Thinking be retired?
GPT-5.2 Thinking will remain available under Legacy Models in ChatGPT for three months after GPT-5.4's release, with planned retirement on June 5, 2026. Paid users can still select it manually from the model picker during this period.
Does GPT-5.4 support the 1M token context window in ChatGPT?
No. The 1M context window is available in Codex (experimental) and the API, not in standard ChatGPT. Context windows in ChatGPT for GPT-5.4 Thinking remain unchanged from GPT-5.2 Thinking. In Codex, the extended window requires configuring model_context_window and model_auto_compact_token_limit, and usage beyond 272K tokens counts at 2× the normal rate.
