Google DeepMind Releases Gemma 4 Open Models

#1 𝕏

Google DeepMind launched Gemma 4, a family of Apache 2.0–licensed open models you can run on your own hardware for advanced reasoning and agentic workflows.

Also covered by: @Sebastian Raschka, @Simon Willison, @Philipp Schmid, @Jeff Dean, @Google DeepMind, @Demis Hassabis, @Demis Hassabis, @Sebastian Raschka

#2 𝕏

Qwen unveiled Qwen3.6-Plus, a next-gen multimodal agentic model with smarter, faster coding execution, sharper vision reasoning and a 1M-token context window by default via API, all while maintaining top-tier general performance.

#3 𝕏

Mustafa Suleyman announced three new MAI models—Artemis (75B), Sentinel (150B) and Aegis (200B)—now in Azure AI Foundry, each optimized for multimodal reasoning, real-time data ingestion and retrieval-augmented generation.

#4 📝 OpenAI News

Codex now offers pay-as-you-go pricing for teams - Codex introduced a new pay-as-you-go pricing option aimed at teams. The update provides more flexible billing to better match team usage patterns.

#5 📝 OpenAI News

OpenAI acquires TBPN - OpenAI announced the acquisition of TBPN. The post briefly describes the acquisition and its strategic intent for the company.

#6 𝕏

OpenAI launched ChatGPT voice mode in CarPlay, bringing hands-free AI assistance to the road. It’s rolling out now to iPhone users on iOS 26.4+ with CarPlay support.

#7 📝 Simon Willison

Gemma 4: Byte for byte, the most capable open models - Notes on Google's Gemma 4 family: four new vision-capable Apache 2.0 licensed reasoning LLMs with emphasis on parameter efficiency and multimodal capabilities including vision and audio. Simon reports on trying the GGUFs locally, model sizes' behavior, and providing example pelican SVG outputs.

Also covered by: @Sebastian Raschka, @Simon Willison, @Philipp Schmid, @Jeff Dean, @Google DeepMind, @Demis Hassabis, @Demis Hassabis, @Sebastian Raschka

#8 ▶️

Why AI came for coders first, automation timelines, and how we’re inside the AI inflection

Lennys Podcast

#9 in

Colin Matthews reports that only ~20 of 51 teams importing design systems into AI prototyping tools use Figma as their source of truth, with the remainder on GitHub, Storybook or MCP.

#10 𝕏

LlamaIndex 🦙 introduces Extract v2 with simplified tiers, pre-saved extraction configurations, and fully configurable document parsing for more powerful, streamlined data extraction. Extract v1 remains accessible under Settings → General.

#11 𝕏

Harrison Chase showcases how @vishsuresh_ built an automated feedback loop for their GTM agent, with step-by-step implementation details available in the linked blog post.

#12 𝕏

Santiago Pika now lets you send Google Meet invites to agents like OpenClaw or Claude so they can join live video calls, enabling true “face-to-face” AI interactions. He thinks this seamless integration is pure gold.

#13 𝕏

Claude launched computer use support in Claude Cowork and Claude Code Desktop for Windows users.

#14 𝕏

Anthropic used representational probes to map one of its recent LLM’s hidden states onto core emotion concepts—tracking how valence, arousal, and discrete emotions emerge across layers.

#15 𝕏

Cursor launched Composer 2, a frontier model with high usage limits, then gave agents their own cloud PCs for true autonomy. Now with Cursor 3, they’re rolling out a new interface to collaborate seamlessly with agents on software.

#16 𝕏

Harrison Chase reports that open-source models now excel at file operations, summarization, tool use, and retrieval—capabilities strong enough to power agent frameworks like Deep Agents.

#17 𝕏

Mustafa Suleyman launched three Microsoft AI models in just months: MAI-Transcribe-1 (world’s most accurate transcription across 25 languages per FLEURS WER), MAI-Voice-1 for ultra-natural speech, and MAI-Image-2 (top 3 model on Arena).

#18 in

Russell Bradley-Cook ⚡️ HubSpot’s ecosystem ranked in the global top 10 alongside AWS, Microsoft & Salesforce in Partnership Leaders’ 2026 Ecosystem Compass report. He underscores partnerships as a vital competitive moat in AI.

#19 in

Dharmesh Shah showcases HubSpot’s Prospecting Agent (50% credit cost cut on April 14), which targeted 116 closed-lost companies, sent 71 emails, and generated 2 replies and 2 meetings.

#20 in

Michael Kurson argues that HubSpot’s decision to open its platform—leveraging 290,000+ customers, $3 billion+ in revenue, a decade of structured GTM data and 2,000+ app partners—is the blueprint for an autonomous CRM.

#21 𝕏

Cursor launched an update that lets you run unlimited agents everywhere—locally, in a worktree, over SSH, or in the cloud—while surfacing its best editor features on demand.

#22 𝕏

Anthropic mapped out 21 “functional emotion” circuits in GPT-2 (e.g. joy, fear, surprise), showing how these emergent modules drive token-level sentiment, generalize across tasks and languages, and can be intervened on to steer model outputs.

#23 𝕏

Jeff Dean shared benchmark results for multiple in-house models and compared their performance head-to-head with Meta’s Gemma 3.

#24 𝕏

Qwen launched its flagship Qwen3.6-Plus model on Fireworks AI, delivering industry-leading inference speed, cost efficiency, and fine-tuning support on their high-performance serving stack.