Tony Kim
Mar 07, 2026 09:18
OpenAI’s GPT-5.4 rolls out across GitHub Copilot with 1M token context window and autonomous computer control capabilities for enterprise developers.
GitHub has made OpenAI’s GPT-5.4 generally available across its Copilot AI coding assistant, bringing the model’s native computer control capabilities and 1 million token context window to developers on paid tiers.
The rollout, announced March 5, 2026, covers Copilot Pro, Pro+, Business, and Enterprise subscribers. GPT-5.4 can now be selected across Visual Studio Code, Visual Studio, JetBrains IDEs, Xcode, Eclipse, and GitHub’s web and mobile platforms.
What’s Different About GPT-5.4
This isn’t just an incremental upgrade. GPT-5.4 introduces autonomous computer control—the model can simulate mouse and keyboard actions to work across different applications without human intervention. For developers, that means Copilot can now handle multi-step workflows that previously required manual switching between tools.
The 1 million token context window represents a substantial jump from previous versions, letting developers feed entire codebases into conversations without hitting truncation limits. GitHub’s internal testing showed GPT-5.4 hitting “new rates of success” on real-world agentic software development tasks, with notably improved performance on complex, multi-step processes.
OpenAI’s benchmarks back this up: GPT-5.4 shows a 33% reduction in factual errors compared to GPT-5.2, and matched or exceeded industry professionals in 83% of GDPval benchmark comparisons—up from 70.9% for its predecessor.
Access Requirements
Individual Pro and Pro+ users get immediate access through the model picker. Enterprise and Business administrators need to manually enable GPT-5.4 in their Copilot policy settings before team members can use it.
GitHub recommends running the latest IDE versions for optimal performance—older compatible versions will show the model but won’t have fine-tuned prompting parameters.
The Bigger Picture
GPT-5.4 launched the same day in two variants: Thinking (optimized for step-by-step reasoning) and Pro (built for enterprise production workloads). The Thinking variant lets users outline reasoning upfront and intervene mid-response—useful for debugging complex logic chains.
OpenAI has set June 5, 2026 as the retirement date for GPT-5.2 Thinking, giving teams roughly three months to migrate workflows. Developers heavily reliant on Copilot’s agent mode should start testing GPT-5.4 now to identify any prompt adjustments needed before the cutover.
Image source: Shutterstock
