OpenAI releases GPT-5.5

OpenAI/GPT models

OpenAI released GPT-5.5 for multi-step computer work including coding, research, data analysis and operating software autonomously until tasks complete. The model matches GPT-5.4's per-token speed while scoring much higher and uses fewer tokens on the same Codex tasks. API access followed a day later.

• Positioned as agent runtime for planning, calling tools, inspecting outputs and iterating without constant user prompting
• Claims state-of-the-art results across 14 benchmarks including 82.7% on Terminal-Bench 2.0 and 58.6% on SWE-Bench Pro
• Launched to ChatGPT Plus, Pro, Business, Enterprise and Codex users on qualifying plans on April 23, 2026; API access April 24
• Excels at multi-step workflows: breaking complex instructions into smaller steps, executing sequentially, refining based on intermediate results