GPT-5.5: OpenAI's New Model Excels at Agentic Coding and Knowledge Work
GPT-5.5 is here, and OpenAI is calling it their smartest and most intuitive model yet. The new release is built specifically for agentic workflows: code tasks spanning multiple steps, web research, data analysis, document creation, and software operation. For Norwegian businesses evaluating AI tools for coding and knowledge work, this is worth noting.
What's new in GPT-5.5
The headline is that GPT-5.5 performs better than its predecessor without being slower. OpenAI claims the model matches GPT-5.4 per-token latency in real-world serving while delivering measurably higher scores across benchmarks. That's a rare combination. Larger, more capable models tend to be slower, not equally fast.
The model is designed to manage entire workflows, not just individual tasks. Instead of the user guiding every step, you can give GPT-5.5 a complex multi-part task and trust it to plan, use tools, check its own work, navigate ambiguity, and keep going until the job is done. This applies especially to coding, data analysis, web research, and document production.
GPT-5.5 is rolling out to Plus, Pro, Business, and Enterprise users in ChatGPT and Codex. GPT-5.5 Pro, designed for even higher-precision work, is coming to Pro, Business, and Enterprise. API access is arriving "very soon," according to OpenAI. Free users are not getting access this time.
API pricing is set at $5 per million input tokens and $30 per million output tokens. That's double GPT-5.4, which cost $2.50/$15.
The benchmarks
OpenAI has published results across a range of established tests. The table below shows GPT-5.5 against its predecessor GPT-5.4 and competitor Claude Opus 4.7 from Anthropic.
| Benchmark | GPT-5.5 | GPT-5.4 | Claude Opus 4.7 |
|---|---|---|---|
| Terminal-Bench 2.0 | 82.7 % | 75.1 % | 69.4 % |
| GDPval | 84.9 % | 83.0 % | 80.3 % |
| OSWorld-Verified | 78.7 % | 75.0 % | 78.0 % |
| Expert-SWE (internal) | 73.1 % | 68.5 % | Not reported |
Terminal-Bench 2.0 deserves special attention. This benchmark tests how models handle complex command-line workflows requiring planning, iteration, and tool coordination. GPT-5.5 scores 82.7 percent. Claude Opus 4.7 lands at 69.4 percent. That's not a marginal lead.
GDPval tests knowledge work across 44 real occupations, from finance to legal research to product management. GPT-5.5 matches or beats industry professionals in 84.9 percent of comparisons. That's a high number for a general AI model.
On SWE-Bench Pro, which evaluates real-world GitHub issue resolution, GPT-5.5 reaches 58.6 percent.
One NVIDIA engineer described losing access to the model as akin to losing a limb. When an engineer who uses AI tools professionally every day describes losing access as physical amputation, it says something about the model's capacity in practice.
What this means for Norwegian businesses
For Norwegian SMBs with 7-100 employees, agentic coding is the most immediate gain. GPT-5.5 can handle longer coding tasks without the user guiding every step. A developer can give the model a feature description and let it build, test, and debug across multiple rounds without intervention. That changes how much a developer can get done per day.
The knowledge work portion is equally relevant. The GDPval score of 84.9 percent means the model performs at the level of professionals across a wide range of disciplines. For Norwegian companies doing analysis, reporting, market research, or product development, it's a real workforce multiplier.
The Codex environment, where GPT-5.5 is particularly strong, is already in use at many development shops. Dan Shipper, founder and CEO at Every, described it as the first coding model with serious conceptual clarity. That's a quote worth noting.
Concrete examples of how Norwegian developers can use GPT-5.5 in practice: A Norwegian fintech company can let a developer describe a new payment flow and have GPT-5.5 build, test, and document the functionality across multiple iterations. An accounting firm can use the model to analyze and format spreadsheet data, or set up automated reports based on structured data. A product team at a Norwegian e-health vendor can have the model conduct research across patient journal systems and produce summaries that take hours instead of minutes.
Cost assessment for a small Norwegian business: A developer actively using Codex with GPT-5.5 for 2-3 hours daily can estimate around 500,000 to 1 million input tokens and 200,000 to 400,000 output tokens per month. Based on $5/$30 per million tokens, that gives a monthly cost of approximately 3,300 to 7,000 Norwegian kroner. This must be weighed against the cost of an hour of developer work: if the company saves 10-20 hours per month, the investment typically breaks even or better.
The price is higher than the predecessor, though. $30 per million output tokens isn't cheap. For a company using the model heavily in coding tools, the monthly cost could be noticeably higher than with GPT-5.4. That has to be weighed against the efficiency gain.
Should Norwegian SMBs upgrade?
For businesses with 1-5 employees and limited technical budget: GPT-5.4 is still a strong choice. Set up a basic workflow first, and consider upgrading once usage is established and budget allows. Start with ChatGPT Plus (249 kr/month) to test the concepts.
For businesses with 5-20 employees and active developers: Upgrading to GPT-5.5 via Codex is recommended. The agentic capacity across long coding tasks justifies the additional cost for most development teams. The investment is measured in saved developer hours, not in subscription cost alone.
For businesses with 20+ employees and complex codebases: GPT-5.5 should be considered as part of a broader AI strategy. The model is suited for teams working with large codebases, complex integrations, or regulatory requirements where precision and documentation matter. In such environments, GPT-5.5's token efficiency is a competitive advantage.
Regardless of company size: GPT-5.5 represents a step up in what AI can be used for in a workflow. For Norwegian SMBs waiting for the "right time" to take AI seriously, this is a time worth noting.
Learn how AIKI can help Norwegian businesses integrate GPT-5.5 into their workflows: AI-kickstart or AI-partner.
- Slug: gpt-5-5-agentisk-ai-koding-norsk-smb
- Kategori: ai-automatisering
- Tags: GPT-5.5, OpenAI, AI-koding, agentisk AI, norske SMB
- Description (NO): GPT-5.5 er OpenAIs nye modell spesielt designet for agentisk koding og kunnskapsarbeid. Les hva den betyr for norske SMB-er.
- Description (EN): GPT-5.5 is OpenAI's new model built for agentic coding and knowledge work. Here's what it means for Norwegian SMBs.
- Title (NO): GPT-5.5: OpenAIs nye modell er spesielt sterk på agentisk koding og kunnskapsarbeid
- Title (EN): GPT-5.5: OpenAI's New Model Excels at Agentic Coding and Knowledge Work
- Foreslått H2 for "Hva betyr dette for norske bedrifter?": Ja
