OpenAI released GPT-5.5 on Thursday, April 23, 2026, its most capable frontier model to date. The release — internally codenamed 'Spud' — comes exactly one month after pretraining wrapped on March 24 and ends weeks of leak-driven speculation after an unreleased OpenAI model was caught live-testing on public APIs earlier in the month.
GPT-5.5 is rolling out immediately to paid ChatGPT and Codex subscribers on the Plus, Pro, Business and Enterprise tiers. API access is coming shortly, pending the integration of additional cybersecurity guardrails that OpenAI says it wants in place before opening the model up to developer traffic.
A new top of the leaderboard
GPT-5.5 debuts at the top of the Artificial Analysis Intelligence Index with a score of 60 — three points ahead of Claude Opus 4.7 and Gemini 3.1 Pro Preview, which had been tied at 57. The lead is narrow on paper but meaningful competitively: Anthropic has held or shared the pole position across much of the last two quarters, and the index is one of the closer-watched composite measures in the industry.
OpenAI's pitch, however, is about agentic capability rather than trivia-style benchmarks. On Artificial Analysis's Coding Agent Index — which measures end-to-end performance on real coding tasks — GPT-5.5 posts state-of-the-art numbers at roughly half the cost of competing frontier coding models. The company is framing the model as one that can take a messy, multi-part request and 'plan, use tools, check its work, navigate through ambiguity, and keep going' with minimal user prompting.
Benchmark highlights
OpenAI published a cluster of headline results alongside the launch:
- Terminal-Bench 2.0: 82.7%, a state-of-the-art score on a benchmark that tests complex command-line workflows requiring planning, iteration and tool coordination.
- SWE-Bench Pro: 58.6% on real-world GitHub issue resolution, a single-pass result that outpaces previous GPT-5.x models.
- OSWorld-Verified: 78.7% on the benchmark that measures whether a model can operate an actual computer environment end-to-end.
- GDPval: 84.9% across 44 occupations' worth of well-specified knowledge work.
Speed has historically been the trade-off as frontier models get bigger, but OpenAI says GPT-5.5 matches GPT-5.4's per-token latency in real-world serving while operating at a substantially higher level of capability.
Why it matters
GPT-5.5 raises the bar on three fronts at once: computer use, autonomous coding, and knowledge work across tools. Those are precisely the axes that matter for the 'agent' products enterprises are now rolling out — Codex-style coding agents, browsing agents, and workspace assistants that have to stay on task across dozens of steps. A coding-agent SOTA at half the cost of the competition is likely to pressure Anthropic and Google on pricing as much as on capability.
It also resets expectations for what a point-release cadence means at OpenAI. GPT-5 launched less than a year ago; since then the company has shipped 5.2, 5.3, 5.4 and now 5.5, with Codex variants layered on top. The platform story — models plus Codex plus native computer use — is clearly the product now, and GPT-5.5 is the clearest expression of it yet.
