OpenAI Launches GPT-5.4 and Acquires Promptfoo to Lock Down Agentic AI

GPT-5.4: OpenAI's Most Capable Model Yet

OpenAI released GPT-5.4 on March 5, 2026, marking a major leap in both capability and cost-efficiency. Available in two variants — GPT-5.4 Thinking (for reasoning-heavy tasks) and GPT-5.4 Pro (maximum performance) — the model introduces several firsts for OpenAI.

The headline feature is a 1 million token context window, matching Anthropic's Claude Opus 4.6 and Google's Gemini 3.1 Pro. But the real differentiator is price: at $30 per million output tokens, GPT-5.4 costs less than half of Claude Opus 4.6's $75/M — a deliberate move to win enterprise budgets.

Computer Use API: Screens, Clicks, and Keystrokes

GPT-5.4 ships with a Computer Use API — a new interface enabling the model to see screens, move cursors, click elements, type text, and interact with desktop applications programmatically. This puts OpenAI in direct competition with Anthropic's computer use capabilities, signaling that screen-level AI interaction is becoming table stakes for frontier models.

Five Levels of Reasoning on Demand

GPT-5.4 introduces configurable reasoning effort with five discrete levels: none, low, medium, high, and xhigh. Developers can control how deeply the model thinks per request, optimizing the cost-quality tradeoff for each use case. This mirrors the adaptive thinking approach pioneered by Claude and Gemini's Deep Think, but with more granular control.

Benchmark Performance

The numbers back up the claims. GPT-5.4 achieves approximately 80% on SWE-bench Verified, placing it in direct competition with Claude Opus 4.6 for autonomous coding. It's 33% less likely to make factual errors compared to GPT-5.2, and scores 87.3% on internal spreadsheet modeling tasks — up from 68.4%.

OpenAI Acquires Promptfoo for Agent Security

On March 10, OpenAI announced its acquisition of Promptfoo, the open-source AI testing startup used by over 25% of Fortune 500 companies. Promptfoo's red-teaming and evaluation tools will be integrated into OpenAI Frontier, the company's enterprise agent platform.

This acquisition fills a critical gap: as agentic AI systems handle more autonomous tasks, the attack surface expands dramatically. Promptfoo's technology enables automated red-teaming, security evaluation of agentic workflows, and compliance monitoring — capabilities that enterprise customers demand before deploying autonomous agents at scale.

Promptfoo has raised $23M to date, including an $18.4M round led by Insight Partners with participation from Andreessen Horowitz. OpenAI confirmed Promptfoo will remain open source.

GPT-5.1 Retirement

As part of the rollout, OpenAI retired GPT-5.1 models on March 11, auto-migrating all conversations to GPT-5.3 Instant, GPT-5.4 Thinking, or GPT-5.4 Pro. The consolidation simplifies OpenAI's model lineup and pushes users toward its latest capabilities.

Why This Matters

GPT-5.4 represents OpenAI's answer to the multi-front competition from Anthropic, Google, and DeepSeek. By combining frontier capabilities with aggressive pricing and strategic acquisitions, OpenAI is positioning itself not just as a model provider but as a full-stack platform for autonomous AI workflows.

—

Further reading: compare model pricing in our LLM pricing calculator and explore the AI tools directory for development workflows.

TL;DR

GPT-5.4 launches with 1M context, Computer Use API, and 5-level configurable reasoning at $30/M output tokens — half the cost of Claude Opus 4.6. OpenAI acquires Promptfoo to secure its agent platform, and retires GPT-5.1 models.