ENFR
8news

Tech • IA • Crypto

TodayMy briefingVideosTop articles 24hArchivesFavoritesMy topics

Gemini 3.5 Flash, OpenAI Dreaming, Majorana 2 reshape AI

AISaturday, June 6, 2026· 12 videos

Briefing

Audio player
0:00 / 0:00

Gemini 3.5 Flash reshapes pricing

Google’s Gemini 3.5 Flash signals the end of near-unlimited AI usage as token limits and tiered subscriptions become standard. Plans like AI Pro, Ultra (€99/month), and Ultra Premium (€219/month) introduce strict caps and upsells for multimodal generation. This forces users to actively manage consumption across Gemini, ChatGPT, and Claude. The shift reframes AI as a metered resource rather than an always-on utility.

OpenAI pushes autonomous Phase 3 AI

OpenAI is advancing toward “Phase 3” systems that act proactively without explicit prompts. These agents aim to manage workflows, emails, and decisions by integrating with user tools and data. The strategy reduces reliance on prompting skills but increases system opacity. It marks a transition from assistants to continuously operating autonomous software.

Dreaming memory transforms ChatGPT usage

OpenAI’s Dreaming introduces persistent, evolving memory that updates continuously in the background. The system refines user profiles over time, enabling highly personalized and anticipatory responses. It can incorporate signals beyond chats, including external tools and contextual data. However, concerns are rising about data retention and the difficulty of fully deleting sensitive information.

Microsoft Majorana 2 targets 2029

Microsoft’s Majorana 2 chip could accelerate practical quantum computing to 2029, ahead of prior estimates near 2033. The redesign replaces aluminum with lead-based superconductors, improving stability. Qubit coherence has reportedly jumped from microseconds to around 20 seconds, enabling progress toward logical qubits. This marks a కీల shift from experimental systems to potentially usable machines.

Codex Sites enables instant apps

OpenAI Codex Sites allows teams to build and deploy internal applications directly from prompts. Each app includes built-in hosting, authentication, storage, and database infrastructure. Companies are replacing documents and dashboards with interactive tools like Account Lens and Compass Memo. The approach compresses development cycles from weeks to minutes.

Anthropic Claude writes 80% code

Anthropic reports that over 80% of its codebase is now generated by Claude, up sharply from 2025. Engineering teams have increased output 8×, while model success rates rose to 76% on complex tasks. In one case, Claude resolved a critical system failure in two hours versus days for humans. The trend suggests early signs of AI accelerating its own development.

GMA 4 12B runs locally

GMA 4 12B is an open multimodal model capable of running on consumer hardware with around 16 GB RAM. It supports text, image, audio, and video natively without separate pipelines. Despite its size, it reportedly handles ~90% of common tasks effectively. This strengthens the case for local, sovereign AI as an alternative to cloud dependence.

MCP automates CRM with Claude

Model Context Protocol (MCP) integrations with tools like GoHighLevel enable AI agents to directly operate CRM systems. Businesses report saving 10 hours weekly and boosting revenue by up to $10,000/month through automation. Use cases include lead triage, pipeline updates, and automated communication. This highlights AI’s growing role as an operational layer inside business software.

Videos covered

Previous briefings · AI