Two weeks after the launch of GPT-5.6 (9 July) and three days after Claude Opus 5 (24 July), the picture is clear enough for a fair cost comparison. Both models are frontier-class, but the pricing structure, EU data routing and strengths differ enough to make the choice sector-dependent.
Prices side by side
OpenAI launched GPT-5.6 as a three-tier family:
| Model | Input (per M tokens) | Output (per M tokens) |
|---|---|---|
| GPT-5.6 Luna | $1 | $6 |
| GPT-5.6 Terra | $2,50 | $15 |
| GPT-5.6 Sol | $5 | $30 |
Claude Opus 5 sits in the same price bracket as its predecessor: $5 per million input tokens and $25 per million output tokens, slightly cheaper on output than GPT-5.6 Sol. Fast mode (≈2.5× faster) costs $10/$50 per MTok.
On benchmarks, Opus 5 clearly leads on ARC-AGI-3 (30.2% vs. 7.8% for Sol) and OSWorld 2.0 (70.6% vs. 62.6%), while GPT-5.6 Sol wins on DeepSWE (72.7%), the benchmark for software engineering tasks. In everyday use, many users find both models almost equivalent, reports Simon Willison.
Where the data goes
For Dutch professionals, technical performance is only half the story. The other half is data processing.
Claude (API and Enterprise): Anthropic offers a GDPR data processing agreement (DPA) for Team and Enterprise customers. Training data: prompt data is by default not used for training via the API and business plans. EU data residency is available via AWS Bedrock or Google Cloud Vertex AI in EU regions, not directly via claude.ai. Consumer accounts (claude.ai free/Pro) are not suitable for processing personal data on behalf of an employer.
GPT-5.6 (API and Enterprise): OpenAI also offers a DPA and EU data residency for Enterprise, Edu and API customers (available since February 2026). Luna and Terra are good for everyday professional tasks and considerably cheaper than Sol or Opus 5.
Both providers meet the minimum requirements for GDPR-compliant business API use. The difference lies in the details of your data processing agreement and the region setting, so always check these in your dashboard.
What suits which sector?
- Legal, healthcare, HR (lots of personal data): Opt explicitly for API or Enterprise with EU data residency enabled. GPT-5.6 Terra or Opus 5 via Bedrock EU are then solid options. See our safety profile per model for an overview.
- Software development and coding: GPT-5.6 Sol scores best on DeepSWE; Opus 5 wins on computer use and agentic tasks. For pure code generation, GPT-5.6 Terra is also a sharp alternative at half the price of Sol.
- Content, analysis, research: Opus 5 has a context window of 1 million tokens and a knowledge cut-off of May 2026, the most recent of the two. GPT-5.6 Sol has the same 1.0.5M-token window. For long documents, that makes little difference.
- Cost optimisation: GPT-5.6 Luna ($1/$6) is the cheapest option for tasks where top-1 quality is not critical. A hybrid approach, with Luna for bulk work and Sol or Opus 5 for complex reasoning, is a common architectural choice.
What changes on 2 August?
Five days. That is how much time you have left before the AI Act transparency obligation takes effect. From then on, chatbots and virtual assistants must identify themselves as AI at the start of every conversation, and AI-generated content must be recognisable as such. This applies regardless of which model you use and regardless of the size of your organisation.
See the model overview or the sector pages for a comparison by sector.
Sources
- OpenAI: GPT-5.6 announcement
- Anthropic / MarkTechPost: Claude Opus 5 release
- kie.ai: Claude Opus 5 vs GPT-5.6 benchmarks
- Simon Willison: GPT-5.6 overview
- Requesty: Claude EU data residency via Bedrock/Vertex
- OpenAI: EU data residency
- Legalz: AI Act transparency obligation 2 August