On 13 August 2026, two new models arrived that reshuffle price relationships in the market: Google launched Gemini 3.7 Flash, and DeepSeek moved its V4 Pro model from preview to general availability. Both are relevant to Dutch professionals, but in very different ways.
Gemini 3.7 Flash: halved price, strong coding performance
Google unveiled Gemini 3.7 Flash on 13 August as the successor to Gemini 3.6 Flash, just three weeks after its predecessor, illustrating the rapid pace of Google's Flash series. The model accepts text, images, audio and video with a context window of 1 million tokens and generates up to 64,000 output tokens per request.
What stands out in the performance figures: on FrontierCode 1.1, 3.7 Flash scores 43.6% against 34.4% for its predecessor, and on DeepSWE v1.1 65.3% versus 49.0%. Those are substantial jumps for an interim version. For developers and teams that have a lot of code generated or reviewed, such as software companies, this is a relevant benchmark.
The introductory price is $0.75 per million input tokens and $3.75 per million output tokens, valid until 31 December 2026. After that the price doubles to $1.50 / $7.50. The temporary introductory price is deliberately lower than Gemini 3.6 Flash and also below comparable models from OpenAI and Anthropic. Google AI Studio and Vertex AI offer the model via API; Gemini AI Pro and Ultra subscribers can use it through Gemini Spark.
For the EU context: via Vertex AI, EU data residency is enforceable (Google Cloud offers regional data residency including the Netherlands). That makes this model more usable for business users who want to guarantee GDPR compliance than the consumer app. See the Gemini model page for the full privacy profile.
DeepSeek V4 Pro 0813: GA and an announced price increase
At the same time, DeepSeek released its V4 Pro model from preview (GA launch on 13 August). It uses a mixture-of-experts architecture with 1.6 trillion parameters in total and 49 billion active per token, a context window of 1 million tokens and a maximum output of 384,000 tokens. The model supports both a thinking mode (chain-of-thought) and a faster non-thinking mode.
The current API price is $0.435 per million input tokens (on a cache miss) and $0.87 per million output tokens, which is cheap compared with American competitors. However, DeepSeek has announced a significant price increase for the near future. If you are now making cost-benefit calculations based on the current rates, it is sensible to factor that range into your figures.
GDPR concern: DeepSeek's official cloud service stores data in China, without an EU adequacy decision. For Dutch organisations in regulated sectors, namely healthcare, legal and financial advice, DeepSeek's direct API remains a risk. Alternatives are self-hosting the open weights on EU infrastructure, or use via Microsoft Azure (Foundry, available since May 2026 for the V4 generation), where EU data residency is guaranteed. See our safety page for an overview per model.
What does this mean taken together?
The two launches illustrate a broader pattern: the AI market competes on performance and price at the same time, but the cheapest options are not always the most GDPR-friendly. For Dutch professionals, the trade-off always has three parts: what the model can do, what it costs, and where my data is processed.
Gemini 3.7 Flash offers a reliable EU route via Vertex AI. DeepSeek V4 Pro offers frontier performance at low cost, but requires active choices about infrastructure. Both models are available today through their respective APIs.
Sources: SiliconAngle, 13 Aug 2026 · MarkTechPost, 13 Aug 2026 · Unite.AI, 13 Aug 2026 · DeepSeek V4 Pro news · LLM Gateway timeline