Three developments from the past week that together further define the playing field for Dutch professionals: OpenAI refined its strongest model, Meta has entered the coding tool market, and the EU AI Act moved into a new enforcement phase as of 2 August.
GPT-5.6 Sol: fewer errors, more control for the user
On 6 August, OpenAI rolled out an update to GPT-5.6 Sol, the top variant in the GPT-5.6 family above Terra and Luna. The main change: the model gives shorter, more direct answers and, according to OpenAI's own measurement, contains 68% fewer factual errors than GPT-5.5 Instant. Plus and Pro subscribers also get a slider that lets them set, per question, how much reasoning effort the model uses, helping you keep costs under control for simple tasks.
At the same time, GPT-5.6 Luna (the smallest variant) is now the default model for free users, with unlimited text chats. This makes it easier for Dutch professionals exploring the platform to make a well-founded comparison with paid options.
For a complete picture of the GPT-5.6 family and the accompanying privacy statement for EU data, see our ChatGPT model page.
Meta Muse Code: a coding tool with parallel subagents
On 5 August, Meta launched Muse Code in public beta, a terminal-based coding tool running on the new Muse Spark 1.2 model. The approach differs from existing tools: Muse Code spawns multiple parallel subagents in isolated git worktrees, coordinates them through an async event log, and only applies changes after explicit approval.
The price is $1.25 per million input tokens and $4.25 per million output tokens. There is also an opt-in training discount: if you allow Meta to use session code for training, you pay $0.10 and $0.20. The difference is worth weighing for software and ICT teams, but it naturally also raises data questions.
Meta has not issued specific statements on EU data processing with the beta launch. Until that is clarified, organisations in regulated sectors should exercise the usual caution: do not process personal data or client code via the standard API route until the data processing agreement and data location have been confirmed. See our safety page for an overview of privacy profiles per provider.
AI Act: enforcement started on 2 August
Just as relevant as the new models: the European AI Act entered an active enforcement phase on 2 August. Through Regulation (EU) 2026/1744, the so-called Digital Omnibus on AI, three obligations became enforceable on that date:
- Prohibited practices (Article 5): AI that subliminally influences people, social scoring by governments, or biometric categorisation on sensitive characteristics is prohibited.
- Transparency for GPAI providers (Article 50): Systems such as chatbots and generated images must be clearly identifiable as AI; providers are obliged to enforce this technically.
- Enforcement and fines: National supervisors, in the Netherlands the Rijksinspectie Digitale Infrastructuur and the Autoriteit Persoonsgegevens, may now act actively.
One nuance emphasised by several sources: the deadline for high-risk AI systems (Annex III, such as CV screening and credit assessment) has been postponed to December 2027, according to Niels Roest. That gives organisations more time to implement, but the transparency and prohibition requirements already apply in full.
In practice: if you let customers or visitors communicate with a chatbot, that must now be clearly recognisable as an AI interaction. The same applies to AI-generated content in newsletters, reports or on social media. Lucas IT and CFO.nl outline the steps: inventory which AI systems your organisation uses, record responsibilities, and make sure AI use is visible to end users.
For translation into specific sectors, such as healthcare, legal and HR, the transparency obligation applies directly to existing workflows. Our safety page tracks which models have which processing locations and compliance statuses.
Sources: OpenAI preview GPT-5.6 Sol · DataNorth on the Sol/Luna update · TechCrunch on Muse Code · MarkTechPost on Muse Code · Lucas IT on the AI Act · CFO.nl on the AI Act · Niels Roest on the postponement