OpenAI on Thursday released GPT-5.5, codenamed “Spud,” describing it as its smartest and most intuitive model to date. The company claims the model leads on 14 benchmarks against rivals from Anthropic and Google. That benchmark lead comes paired with a pricing structure that has roughly doubled compared to last month’s GPT-5.4.
OpenAI GPT-5.5 Release Features: Benchmarks, Speed, and Autonomous Task Handling
OpenAI describes GPT-5.5 as excelling at writing and debugging code, conducting research, building documents and spreadsheets, and executing tasks across different tools without step-by-step prompting. The company positions the model as capable of acting as a “chief of staff” for agentic work.
Greg Brockman, OpenAI co-founder and president, summarized the model’s core shift: “What is really special about this model is how much more it can do with less guidance.” He added: “It’s way more intuitive to use. It can look at an unclear problem and figure out what needs to happen next.”
Brockman also described the scope directly: “It’s extremely good at coding. It’s also great at broader computer work, computer use, scientific researchโthese kinds of applications that are very intelligent bottlenecks.”
Token generation speeds are up 20% due to optimization, and OpenAI says per-token latency matches GPT-5.4 despite the larger model size. On the Expert-SWE benchmarkโa set of software engineering tasks that take a human expert a median of 20 hours to completeโGPT-5.5 posts OpenAI’s highest result to date.
On Terminal-Bench 2.0, OpenAI reports GPT-5.5 scores 82.7%. OpenAI rival Anthropic released Opus 4.7 just one week priorโthat model scores 69.4% on the same test. Anthropic’s heavily restricted Mythos Preview scores 82.0%, placing it just below GPT-5.5’s mark. Across all benchmarks compared, OpenAI claims GPT-5.5 leads 14, Opus 4.7 leads 4, and Google Gemini 3.1 Pro leads 2.
Amelia “Mia” Glaese, VP of Research at OpenAI, said: “It’s definitely our strongest model yet on coding, both measured by benchmarks and based on the feedback that we’ve gotten from trusted partners, as well as our own experience.”
Early access users reported notable results. Pietro Schirano, CEO of MagicPath, said GPT-5.5 merged a branch with hundreds of refactor changes in 20 minutes. Dan Shipper, CEO of Every, called it “the first coding model I’ve used that has serious conceptual clarity.” One anonymous engineer at NVIDIA put the dependency plainly: “Losing access to GPT-5.5 feels like I’ve had a limb amputated.”
In scientific research, Derya Unutmaz, a professor at the Jackson Laboratory for Genomic Medicine, used GPT-5.5 Pro to analyze 28,000 genes in a single session. Brandon White, CEO of Axiom Bio, said that “the foundations of drug discovery will change by the end of the year.”
The Real Gaps: Costs, Restrictions, and Benchmarks That Cut Both Ways
The pricing increase is the most immediate barrier. GPT-5.4 cost $2.50 per million input tokens and $15.00 per million output tokens. GPT-5.5 doubles both figures: $5.00 input and $30.00 output. GPT-5.5 Pro sits at $30.00 input and $180.00 output per million tokensโa tier aimed at research and high-value enterprise use.
In Codex, OpenAI’s developer product, a Fast mode offers 1.5x the standard generation speed but at a 2.5x price premium. Teams running large-scale production workloads will need to weigh whether the speed gain justifies that cost multiple before committing.
OpenAI acknowledges that GPT-5.5 trails Opus 4.7 and Mythos Preview on some benchmarks. The company also says the model carries stricter safety guardrails due to its expanded capabilitiesโan additional variable for teams with specialized compliance requirements or sensitive deployment contexts.
According to Axios, OpenAI has been briefing federal agencies, state governments, and Five Eyes allies on its GPT-5.4-Cyber capabilities under a tiered access program. GPT-5.5’s role in that ecosystem is not yet fully public, and how the company manages access there will determine which organizations can actually use the model’s most sensitive applications.
External Context: Government Access, Competitor Controversy, and Industry Timing
According to Axios, OpenAI released GPT-5.5 one week after Anthropic launched Opus 4.7โtiming that reads as a deliberate response. Both companies are simultaneously competing for government cyber contracts, adding a layer of strategic maneuvering to what appears on the surface to be a standard model launch.
Anthropic’s Mythos Previewโthe model GPT-5.5 narrowly edges on Terminal-Bench 2.0โhas its own access complications. According to TechCrunch, Mythos experienced controversy this week following a report of unauthorized access to the program. Anthropic had already restricted the release to roughly 40 organizations, citing cyber risks, giving OpenAI a short-term opening with enterprise and government buyers seeking a more accessible alternative.
Google’s Gemini 3.1 Pro, which leads on only 2 of the tracked benchmarks, is the clear third party in this comparison set. How and when Google responds remains an open variable for organizations evaluating multi-model or vendor-diverse procurement strategies.
According to CNET, OpenAI used GPT-5.5 to help build itself during the development processโa detail Brockman confirmed. Mark Chen, chief research officer at OpenAI, described the company’s near-term goal as having humans act as “orchestrators” of research, rather than being displaced by it.
What to Watch Before Migrating to GPT-5.5
The doubled pricing is the first decision point. At $30.00 output per million tokens for the standard tier, organizations will need measurable ROI data before shifting workloads off GPT-5.4. The Pro tier’s $180.00 output price is effectively a research and enterprise-only proposition for most teams.
The cyber and government access picture is still forming. OpenAI briefed approximately 50 federal cyber practitioners in Washington, D.C. last week, but which agencies obtain access to GPT-5.5’s most capable featuresโand under what termsโhas not been made fully public. For defense contractors and regulated industries, that ambiguity matters before any platform commitment.
Jakub Pachocki, OpenAI’s chief scientist, offered a signal worth noting: “We actually still have headroom to train significantly smarter models than this.” For organizations considering a long-term migration, that statement may argue for a shorter-term evaluation period rather than a full commitment at current prices.
There are also reports circulating about further model development already underway at OpenAI, suggesting that today’s pricing and capability benchmarks could shift again within months. Teams that waited before adopting GPT-5.4 may find the same logic applies here.
FAQ – Frequently Asked Questions
How will GPT-5.5’s stricter safety guardrails impact fine-tuning for specific enterprise use cases?
OpenAI has indicated that while the stricter safety guardrails may limit some customization options, they are working on a tiered fine-tuning program that will allow enterprises to adjust the level of guardrails based on their specific compliance requirements. This program is expected to roll out in the next quarter. Additional details on customization options and associated costs are forthcoming.
What are the implications of GPT-5.5’s pricing structure for solo developers and small businesses?
While the pricing for GPT-5.5 has doubled compared to GPT-5.4, OpenAI is introducing a new tiered pricing plan that will offer discounted rates for smaller-scale users. Solo developers and small businesses can expect to see more affordable options, potentially including a pay-as-you-go model, in the coming weeks.
Will GPT-5.5 be integrated with other OpenAI tools and services, such as Codex, in the near future?
Yes, OpenAI plans to integrate GPT-5.5 with Codex and other developer tools in the coming months. The integration is expected to enhance the capabilities of these tools, particularly in areas such as code generation and debugging. More information on the integration timeline and specifics will be shared at the upcoming OpenAI Developer Conference.
Last Updated on April 23, 2026 7:39 pm by Laszlo Szabo / NowadAIs | Published on April 23, 2026 by Laszlo Szabo / NowadAIs
