OpenAI Launches GPT-5.5 and GPT-5.5 Pro: What to Know

OpenAI has released GPT-5.5 and GPT-5.5 Pro, calling the pair its smartest and most intuitive models yet. Both became available in the API on April 24, 2026, alongside a rollout across ChatGPT and Codex.

What's new

GPT-5.5 is pitched as a model built for complex, real-world work — writing code, researching online, analyzing information, drafting documents and spreadsheets, and moving across tools. According to OpenAI, relative to earlier models it understands tasks earlier, asks for less guidance, uses tools more effectively, checks its own work, and keeps going until a task is complete.

GPT-5.5 Pro is not a separate model — it's the same underlying system running with a setting that applies parallel test-time compute, effectively giving it more "thinking" resources for harder problems.

Availability differs by product:

GPT-5.5 is rolling out to Plus, Pro, Business, and Enterprise users in ChatGPT and Codex.
GPT-5.5 Pro is rolling out to Pro, Business, and Enterprise users in ChatGPT only.

No pricing details or tier-specific rollout dates were disclosed — OpenAI describes availability only as "rolling out."

The numbers OpenAI is citing

82.7% accuracy on Terminal-Bench 2.0
58.6% on SWE-Bench Pro
Feedback gathered from nearly 200 early-access partners ahead of release
More than 85% of OpenAI staff reportedly use Codex weekly

These figures come from OpenAI's own reporting. There is no independent or third-party benchmark verification included in the release, and no side-by-side comparison showing how GPT-5.5 performs against rival models on the same tasks.

Early reactions

OpenAI shared testimonials from early users:

Dan Shipper, Founder and CEO of Every, called GPT-5.5 "the first coding model I've used that has serious conceptual clarity."
Pietro Schirano, CEO of MagicPath, said it "genuinely feels like I'm working with a higher intelligence, and there's almost a sense of respect."
An NVIDIA engineer said, "Losing access to GPT-5.5 feels like I've had a limb amputated."

These are testimonials from a limited set of early-access partners and may not generalize to broader real-world use.

Safety and safeguards

OpenAI says GPT-5.5 went through a full suite of predeployment safety evaluations under its Preparedness Framework, including targeted red-teaming for advanced cybersecurity and biology capabilities. The company describes this as the strongest set of safeguards to date, intended to reduce misuse while preserving legitimate uses. The system card was updated on release day with additional safeguard details specific to API deployment.

No specifics were given on exactly what the new safeguards entail or how "strongest to date" is measured — the targeted red-teaming for cyber and bio capabilities does suggest the model may carry elevated misuse risk that requires ongoing management.

Where this fits in the GPT-5 lineup

The release adds to a fast-moving GPT-5 series. GPT-5.3 Instant, the most recent addition before GPT-5.5, responds faster and reduces dead ends, unnecessary caveats, and overly declarative phrasing in its answers. GPT-5.4 Thinking, described as the latest reasoning model in the line, is the first general-purpose OpenAI model with mitigations for High capability in cybersecurity, building on approaches used in GPT-5.3 Codex, ChatGPT, and the API; it's baselined against GPT-5.2 Thinking. Notably, there is no model named GPT-5.3 Thinking. OpenAI did not provide dates for the GPT-5.3 Instant or GPT-5.4 Thinking releases, so it's unclear exactly how they line up on the timeline relative to GPT-5.5.

For founders building on the API, this rapid version churn — 5.3 Instant, 5.4 Thinking, 5.5 — could create integration or compatibility questions as teams decide which variant to standardize on.

Competitive landscape

OpenAI lists Claude Opus 4.7 and Gemini 3.1 Pro as competing models, but the report includes no benchmark comparisons showing how GPT-5.5 stacks up against either on shared tasks.

Why founders should care

If the benchmark gains hold up under independent testing, coding and agentic-task reliability could plausibly improve, which may reduce engineering overhead for teams building products on top of GPT-5.5 — though this remains unverified outside OpenAI's own figures.
Availability across Plus through Enterprise tiers could lower the barrier for startups of different sizes to access the more advanced GPT-5.5 Pro configuration, particularly for complex or high-stakes tasks.
The added safeguards and cyber/bio red-teaming suggest founders integrating GPT-5.5 into products may need to review updated usage policies and the revised system card before deployment.
Given competing models like Claude Opus 4.7 and Gemini 3.1 Pro, founders would be well-served to benchmark across providers on their own workloads rather than relying solely on OpenAI's self-reported numbers before committing to production use.

Bottom line

GPT-5.5 and GPT-5.5 Pro represent OpenAI's latest and, by its own account, most capable release — with real benchmark gains, an expanding safety framework, and strong early testimonials. But with no independent verification, no cross-model benchmarking, and a fast-splintering GPT-5 lineup, founders should treat the claims as a starting point for evaluation rather than a final verdict.