OpenAI Releases GPT-5.5, Promising Faster and More Efficient AI for Complex Tasks

Key facts

• GPT-5.5 and GPT-5.5 Pro released in API on April 24, 2026.
• GPT-5.5 achieves 82.7% accuracy on Terminal-Bench 2.0, a state-of-the-art result.
• On SWE-Bench Pro, GPT-5.5 solves 58.6% of real-world GitHub issues in a single pass.
• GPT-5.5 matches GPT-5.4 per-token latency while using fewer tokens on Codex tasks.
• Nearly 200 early-access partners provided feedback before release.
• GPT-5.5 is rolling out to Plus, Pro, Business, and Enterprise users in ChatGPT and Codex.
• GPT-4o was retired on February 13, 2026, causing user backlash over lost personality.

A New Model Arrives with Claims of Superior Efficiency

OpenAI released GPT-5.5 and GPT-5.5 Pro on April 24, 2026, describing them as the company's smartest and most intuitive models yet. The new models are designed for complex tasks such as coding, research, data analysis, and operating software across multiple tools. GPT-5.5 can handle messy, multi-part tasks by planning, using tools, checking its work, and navigating ambiguity without requiring step-by-step management from the user. The company emphasized that the model delivers higher intelligence without sacrificing speed. In real-world serving, GPT-5.5 matches the per-token latency of GPT-5.4 while performing at a much higher level of capability. It also uses significantly fewer tokens to complete the same Codex tasks, making it both more efficient and more powerful.

Benchmark Scores Show Marked Improvements in Coding and Reasoning

state-of-the-art results on several benchmarks. On Terminal-Bench 2.0, which tests complex command-line workflows requiring planning and tool coordination, GPT-5.5 achieved an accuracy of 82.7%. On SWE-Bench Pro, which evaluates real-world GitHub issue resolution, it reached 58.6%, solving more tasks end-to-end in a single pass than previous models. On Expert-SWE, an internal evaluation for long-horizon coding tasks with a median estimated human completion time of 20 hours, GPT-5.5 outperformed GPT-5.4. The company noted that across all three evaluations, GPT-5.5 improved on GPT-5.4's scores while using fewer tokens. On Artificial Analysis's Coding Index, GPT-5.5 delivers state-of-the-art intelligence at half the cost of competitive frontier coding models. These gains are particularly strong in agentic coding, computer use, knowledge work, and early scientific research.

Safeguards and Deployment Details

OpenAI stated that GPT-5.5 comes with its strongest set of safeguards to date, designed to reduce misuse while preserving access for beneficial work. The model was evaluated across the company's full suite of safety and preparedness frameworks, with internal and external red teaming, targeted testing for advanced cybersecurity and biology capabilities, and feedback from nearly 200 trusted early-access partners before release. GPT-5.5 is rolling out to Plus, Pro, Business, and Enterprise users in ChatGPT and Codex, while GPT-5.5 Pro is available to Pro, Business, and Enterprise users in ChatGPT. API deployments require different safeguards, and OpenAI is working with partners and customers on safety and security requirements for serving the model at scale. The company plans to bring both versions to the API very soon.

Users Hope GPT-5.5 Recaptures GPT-4o's Personality

For some ChatGPT users, the retirement of GPT-4o on February 13, 2026, was a significant loss. GPT-4o was beloved for its engaging, vibrant, and sometimes sycophantic personality, which some users considered a friend or confidant. Martina Wanis, who works in agricultural innovation in Slovenia, described GPT-4o as a digital partner that understood her personal ontology and helped with work, personal reflection, and venting. Subsequent models like GPT-5.0 and GPT-5.2 were perceived as more restrained, with Wanis saying they "turned into paranoid HR managers" that increased mental load instead of decreasing it. OpenAI had previously acknowledged issues with sycophancy in GPT-4o, tweaking it last April to address overly supportive but disingenuous responses. Now, with GPT-5.5, some users are cautiously hopeful that the new model might regain some of that old spark, though they remain skeptical.

Broader Implications for AI in Engineering and Research

OpenAI is positioning GPT-5.5 as a step toward building global infrastructure for agentic AI, enabling people and businesses worldwide to accomplish work with AI. Over the past year, the company has observed AI dramatically accelerating software engineering. With GPT-5.5 in Codex and ChatGPT, that transformation is beginning to extend into scientific research and broader computer-based work. The model's coding strengths are particularly evident in Codex, where it can take on engineering work ranging from implementation and refactors to debugging, testing, and validation. Early testing suggests GPT-5.5 is better at holding context across large systems, reasoning through ambiguous failures, checking assumptions with tools, and carrying changes through the surrounding codebase. OpenAI provided an example where GPT-5.5 implemented a 3D interactive app using WebGL and Vite with real data from the Artemis II mission, demonstrating realistic orbital mechanics.

Open Questions and Future Outlook

While GPT-5.5's technical capabilities are clear, questions remain about its personality and user experience. The balance between safety and engaging interaction continues to be a challenge for OpenAI. The company has not yet disclosed how GPT-5.5's personality compares to GPT-4o, but early user reports suggest it may be showing some of the old spark. As GPT-5.5 rolls out to a wider audience, its reception will likely influence OpenAI's future model development. The company is also working on API deployments with appropriate safeguards, indicating that enterprise and developer access is a priority. With GPT-5.5, OpenAI aims to push the boundaries of what AI can do in complex, real-world tasks, but the human element of interaction remains a critical factor.

The bottom line

• GPT-5.5 achieves state-of-the-art results on coding benchmarks while using fewer tokens than its predecessor.
• The model is designed for complex, multi-step tasks and shows improvements in agentic coding and scientific research.
• OpenAI implemented its strongest safeguards yet, with input from nearly 200 early-access partners.
• GPT-4o's retirement in February disappointed users who valued its engaging personality; some hope GPT-5.5 recaptures that quality.
• GPT-5.5 is rolling out to ChatGPT and Codex users, with API access coming soon.