senn-techAI Radar: GPT-5 from OpenAI
GPT-5 is for many companies the first contact with modern AI — via ChatGPT or the API. The strengths: broad knowledge, strong reasoning, good multimodality, and a massive ecosystem of tools and integrations.
Where it shines
For general tasks, text work, research assistance, and rapid prototyping, GPT-5 is a solid standard tool. Its widespread adoption ensures that examples exist for nearly every language, library, and platform. If you build a solution, you usually find libraries, plug-ins, and best practices already in place — which lowers the entry barrier noticeably. The ecosystem is the real moat: feature extensions, tool integrations, and agent frameworks almost always land on the OpenAI API first before they follow elsewhere. Anyone who needs prototypes finished fast benefits from that directly.
Reasoning and What It Costs
GPT-5 continues the o-series: on hard problems, the model can switch on a chain-of-thought path in which it works through intermediate steps before answering. That lifts quality on math, code, and multi-step plans — and at the same time costs latency and tokens. On trivial tasks, reasoning is wasted compute, which is why handling the modes deliberately pays off. For agentic use, it means: shift the model into the deeper mode only when it truly counts, and otherwise stay on the faster path. A simple rule of thumb helps day to day: the clearer the task, the flatter the path; the more decision steps involved, the more the deeper mode earns its keep.
License: Consistently Closed
GPT-5 is proprietary and API-only — no weights, no self-hosting, no local fine-tuning. There is no official open-weight variant, and the architecture stays with OpenAI. If you want to use the model, you use the cloud — with everything that implies for data sovereignty. That sets GPT-5 fundamentally apart from models under MIT or Apache: there you can load weights, adapt them, and run them in your own data center; with GPT-5, every use stays bound to an external service.
The catch for DACH companies
Data goes to OpenAI — processing outside the EU. For internal, confidential, or personal data, this is an issue that cannot be ignored. Enterprise contracts alleviate it but do not fully resolve it. Concretely: zero-retention agreements and EU processing locations reduce the risk, but the fundamental dependency on a US provider remains. For many SMEs that is acceptable on uncritical tasks — not on sensitive inputs. A clear dividing line helps here: anything that is public or internally unproblematic may go through the cloud; anything with customer, HR, or contract reference stays in-house on a self-hosted model.
Our view
GPT-5 is a good tool — one among several. We use it in a technology-neutral way and switch flexibly between providers via an LLM gateway. For sensitive workloads, the answer remains: on-premise, on our own hardware. GPT-5 is the reference others are measured against — but reference is not automatically the right choice for every task.
Further Reading
When is GPT-5 worth it despite the cloud dependency?+
For general tasks, text work, research assistance, and rapid prototyping that can draw on a vast ecosystem of libraries and best practices. The real moat is this reach: feature extensions and agent frameworks almost always land on the OpenAI API first. For non-critical workloads that is a genuine speed advantage — but for sensitive data GPT-5 is not the right choice.
How do you handle data transfer to OpenAI?+
With a clear dividing line: anything that is public or internally unproblematic may go through the cloud; anything with customer, HR, or contract reference stays in-house on a self-hosted model. Zero-retention agreements and EU processing locations mitigate the risk but do not resolve the fundamental dependency on a US provider. We switch flexibly between providers via an LLM gateway.
Can you self-host or adapt GPT-5?+
No. GPT-5 is proprietary and API-only — no weights, no self-hosting, no local fine-tuning. There is no official open-weight variant; the architecture stays with OpenAI. If you want to use the model, you use the cloud, with everything that implies for data sovereignty. That sets GPT-5 fundamentally apart from models under MIT or Apache that can be run on your own hardware.