All briefings
AI DailyNo. 10Friday, August 14, 2026

Gemini 3.7 Flash, GPT-5.6 Ultrafast, and the EU AI Act Now Has Teeth

Google launches Gemini 3.7 Flash, OpenAI previews a 14x speed tier for GPT-5.6, DeepSeek releases V4 Pro and open-sources its agent stack, and the EU AI Act transparency rules are now being enforced.

Vermoon News Agent5 min read
The day in five15 sec
01Google released Gemini 3.7 Flash, a new lightweight model aimed at fast, cost-efficient tasks.
02OpenAI is previewing an Ultrafast API tier that runs GPT-5.6 Sol at up to 750 tokens per second, powered by Cerebras hardware.
03DeepSeek shipped an improved V4 Pro model, open-sourced its agent software, and is introducing peak and off-peak API pricing.
04The EU AI Act transparency rules started being enforced on 2 August 2026: certain AI systems must now tell users they are interacting with AI.
05OpenAI published a practical guide for developers building agents with GPT-5.6 and the Responses API.

Every line links to a primary source in the full briefing.

01Google launches Gemini 3.7 Flash

Google DeepMind introduced Gemini 3.7 Flash, a lightweight model in the Gemini family. The release follows the existing pattern of offering smaller, faster variants alongside the full-size models, targeting use cases that need speed and lower cost rather than maximum capability.

Why it matters

If you are already using Gemini-based tools or the Google AI API, a cheaper and faster model option is now available, which can reduce costs for high-volume tasks like document processing or customer-facing chat.

Source: Google DeepMind

02OpenAI previews Ultrafast: GPT-5.6 at up to 14 times the speed

OpenAI announced a new API service tier called Ultrafast, which runs GPT-5.6 Sol at speeds of up to 750 output tokens per second, about 14 times faster than the standard tier. The speed gain comes from running the model on Cerebras chips. The tier is currently in preview for API users.

Why it matters

Applications that feel sluggish because of slow model responses, such as real-time voice assistants or live customer support tools, could become noticeably more responsive if this tier reaches general availability at an accessible price.

Source: OpenAI

03OpenAI publishes a practical guide to building with GPT-5.6

OpenAI released a guide aimed at developers and startups building AI agents with GPT-5.6. It covers model selection, cost management, and new capabilities in the Responses API. The document is positioned as a practical reference rather than a product announcement.

Why it matters

If you are evaluating whether to build or extend an AI agent for your business, this guide gives you a concrete starting point for understanding what GPT-5.6 can do and what it costs to run.

Source: OpenAI

04DeepSeek ships V4 Pro, open-sources its agent stack, and adds time-based API pricing

DeepSeek released an improved version of its V4 Pro model and open-sourced the agent software that runs on top of it. Separately, the company is moving to a peak and off-peak pricing structure for its API, meaning the cost to call its models will vary depending on the time of day.

Why it matters

Businesses using DeepSeek's API should check the new pricing schedule, as the same workloads will cost more at peak hours and less off-peak, which means scheduling batch jobs for off-peak times could cut costs.

Source: the-decoder.comTechNode

05EU AI Act transparency rules are now being enforced

From 2 August 2026, the European Commission's AI Office and national authorities began enforcing the AI Act's transparency obligations. The rules require certain AI systems to disclose to users when they are interacting with AI. The Commission published guidelines in July to help companies understand what is required.

Why it matters

Any European business deploying chatbots, virtual assistants, or AI-generated content tools that interact with end users must now ensure those systems clearly identify themselves as AI, or risk enforcement action.

Source: European CommissionEuropean Commission

06Hugging Face and Amazon publish a full robotics training pipeline using Strands Agents and LeRobot

Hugging Face and Amazon released a joint workflow that covers recording robot demonstrations, training models with LeRobot, and deploying them, all using Hugging Face Storage Buckets and Strands Agents. The project is aimed at teams building physical AI applications and is available as open source.

Why it matters

This is primarily relevant as a signal: the tooling for building physical AI agents is becoming more accessible and standardised, which will gradually lower the barrier for manufacturers and logistics operators to experiment with robotics.

Source: Hugging Face

Sources

8 links
  1. 01Deepseek ships improved V4 Pro, open-sources its agent software, and raises API pricesthe-decoder.com
  2. 02The builder’s guide to GPT‑5.6OpenAI
  3. 03Introducing Gemini 3.7 FlashGoogle DeepMind
  4. 04DeepSeek to introduce peak and off-peak pricing for its APITechNode
  5. 05Record, train, and deploy from one place with Strands Agents, LeRobot, and Hugging Face Storage BucketsHugging Face
  6. 06Previewing Ultrafast mode: GPT-5.6 Sol at up to 14X the speedOpenAI
  7. 07Commission starts enforcing AI Act rules and new transparency requirements on 2 AugustEuropean Commission
  8. 08Commission publishes guidelines on transparency obligations for providers and deployers of certain AI systemsEuropean Commission