From Product Hunt

Google's new local multimodal model stands out today alongside dev tools for agentic coding and an AI security platform.

Google Gemma 4 12B — Run multimodal AI locally with an encoder-free architecture Processes text, vision, and audio natively without separate encoders, running on just 16GB VRAM. Apache 2.0 licensed, compatible with Ollama, LM Studio, llama.cpp, and vLLM. Benchmarks near the larger 26B MoE variant.
▲ 235 · blog.google
Astra Autonomous Pentest — AI agents that find, validate, and fix every vulnerability Army of offensive pentesters and bounty hunter agents discovers complex chained vulnerabilities, a validator layer drives false positives to near-zero, and AI-fix agents deliver remediation as Cursor, Copilot, or Claude Code prompts.
▲ 318 · getastra.com
Boxes.dev — Run Claude Code and Codex in your own cloud environment Each Claude Code or Codex chat gets its own computer in the cloud. Desktop, CLI, and mobile apps. Uses your existing subscription. Optimized for parallel agent workflows with Slack integration.
▲ 102 · boxes.dev
Keen Code — Context-efficient CLI coding agent built by agents Open-source Go CLI agent with turn memory to keep sessions lean. Maps MCP servers to lazy-loaded Skills instead of stuffing large schemas into context upfront. Full prompt/design trail preserved in the repo.
▲ 113 · github.com
Extella.AI — Agentic platform that self-evolves and builds reusable systems Remembers what works and reuses it across four memory layers: Rules, Concepts, Experts (automations), and KV store. Bring your own LLM, connect any tool. By Day 30 it's a system only you have. macOS, Windows, Linux.
▲ 89 · extella.ai
Sun — Collaborative voice API for agents Voice AI built for multi-speaker interactions, not 1:1 chat. Multi-speaker turn-taking, agent-aware barge-in, multi-agent in one room, and 10x the context window of ChatGPT Realtime and Gemini Live.
▲ 96 · getsun.io
From Reddit

Massive day for Anthropic: SpaceX compute deal, IPO filing, and KPMG alliance in the same week. Plus a Trump AI executive order and DeepSeek's first fundraise.

Anthropic files confidential S-1 for IPO at $965B valuation Filed days after closing a $65B Series H. Revenue run-rate hit $47B in May, up from ~$10B a year prior. On pace for its first profitable quarter. Could list as early as October 2026, joining SpaceX and OpenAI in the blockbuster IPO race.
KPMG deploys Claude to all 276,000 employees across 138 countries Claude embedded into Digital Gateway (KPMG's Azure-based platform) with Cowork and Managed Agents. Tasks that took weeks now take minutes. KPMG Blaze uses Claude Code to help PE portfolio companies modernize legacy IT.
Trump signs AI executive order: voluntary 30-day pre-release to government Companies can voluntarily share frontier AI models with the government for up to 30 days before public release. The NSA determines which models are "covered frontier models." Explicitly not a mandatory licensing or permitting requirement.
DeepSeek raising $7.4B in first-ever funding round at up to $59B valuation Tencent and CATL among backers. Founder Liang Wenfeng contributing ~40% himself. A dramatic shift for the lab behind V3 and R1 that rose to prominence without outside VC. Could close within weeks.
OpenAI upgrades GPT-5.5 Instant, retires GPT-4.5 on June 27 GPT-5.5 Instant becomes the new default ChatGPT model, replacing GPT-5.3 Instant. GPT-4.5 — launched just four months ago as "the most knowledgeable model" — is fully superseded by GPT-5 and GPT-5.3 Codex.
Broadcom drops 12.6% after AI guidance misses expectations Beat on total revenue ($22.2B, +48% YoY) but AI chip revenue of $16B missed the $17.2B estimate. First major AI infrastructure company to disappoint on AI specifically despite record results overall.
From Reddit

xAI enters the video generation arena. Plus an industry overview from a16z.

Grok Imagine Video 1.5 launches, tops AI video arena leaderboard xAI's video model debuted June 4 at #1 on Artificial Analysis Video Arena for image-to-video. Now available on Vercel for easy integration. Musk showcased it with an AI-generated trailer for the Iliad.
a16z: The State of Generative Media 2026 Comprehensive industry report from Andreessen Horowitz. Multimodal AI systems handling text, image, video, and voice together are shifting from novelty to embedded workflow. Video generation led by Seedance 2.0, HappyHorse, Sora 2, and Veo 3.1.
Figma adds Plan mode for AI-powered Make New opt-in mode helps shape direction before AI generation starts. Professional plan admins can now buy additional AI credits pay-as-you-go.
From Reddit

WWDC 2026 is Monday — macOS 27 with a revamped Siri incoming.

macOS 27 compatibility: M-series Macs only All Intel Macs and some early Apple Silicon models dropped. Full compatibility list ahead of Monday's keynote.