Three developments that together are reshaping the market for Dutch professionals: the EU AI Act has moved beyond the paper stage, OpenAI is cutting the price of its lightest model sharply, and Anthropic has delivered a new flagship variant that outperforms its predecessor at the same price.
EU AI Act: national supervisory authorities are now enforcing
From 2 August 2026, the transparency obligations under Article 50 of the EU AI Act apply. This means national market surveillance authorities are actively monitoring, there is no longer a grace period. The concrete requirements: chatbots and other AI systems must inform users that they are interacting with an AI, and AI-generated text with an informative purpose (news articles, annual reports, government communication on social media) must be labelled as such.
Further into the law, some measures have been postponed: the stringent requirements for high-risk AI under Annex III only apply from December 2027, and products under Annex I follow in August 2028. But the basis for enforcement is in place now, and the Dutch Data Protection Authority can stack fines under the AI Act on top of any GDPR fines, these are cumulative.
Practical starting point for any organisation: make an AI inventory. Record which tools you use, for what purpose, and who is responsible for them. Even if you buy a commercial package, the organisation using it remains liable for how it is applied. On our safety page you will find an overview of the GDPR profiles per provider, including which ones offer a data processing agreement.
Sources: Frankwatching, CFO.nl
GPT-5.6 Luna: 80% price cut as of 30 July
On 30 July, OpenAI cut the API price of GPT-5.6 Luna by 80%: from $1 to $0.20 per million input tokens, and from $6 to $1.20 per million output tokens. GPT-5.6 Terra fell by 20%, to $2 and $12 per million tokens.
Luna is the fast, low-cost model in the GPT-5.6 family, suited to classification, routing, summarisation and lightweight real-time applications. OpenAI cites efficiency gains in its infrastructure as the reason; an additional factor is growing competition from Chinese models and other providers, which is adding pressure on prices.
For teams processing large volumes, such as legal document analysis at law firms or transaction screening at accountancy firms, this is a relevant shift. Luna usage also counts for less against the credits of ChatGPT Work and Codex subscriptions; the subscription prices themselves remained unchanged.
Source: gHacks
Claude Opus 5: better scores, same price as Opus 4.8
Anthropic launched Claude Opus 5 on 24 July at $5 per million input tokens and $25 per million output tokens, half the price of Fable 5, the current top model. Impressive benchmark figures: on Frontier-Bench v0.1 (coding), Opus 5 scored 43.3%, compared with 18.7% for Opus 4.8 and 33.7% for Fable 5. On ARC-AGI 3, the model scored three times as high as the next model.
A Fast mode is available at roughly 2.5× the speed, but at double the price. Prompt caching delivers up to 90% cost savings for repeated contexts; batch processing gives a 50% discount.
Claude Opus 5 is offered by default on the Max subscription ($100–$200/month). For software and ICT teams that code a lot or run complex reasoning chains, the price-performance ratio compared with Fable 5 is a serious consideration.
Anthropic has made no new statements on EU data processing with this launch; the existing data processing agreement for API use remains in force. See our Claude model page for the current GDPR profile.
Source: Technology.org
What this summarises
The market is moving towards cheaper, more efficient models while regulation is being enforced more strictly. For Dutch professionals: check which AI tools your organisation uses, who is responsible for their deployment, and whether the transparency obligations of the EU AI Act have already been built into your workflow. The combination of lower model prices and higher compliance expectations makes a structured approach more worthwhile than ever.