## New AI Tools & Model Launches: Week of June 14–20, 2026 ### 🔥 What Changed This Week This was one of the densest AI launch weeks of 2026, with a convergence of **new foundation models, agentic tools, and product-level AI integrations** across all major players. Here are the headline shifts: --- ### 1. 🧠 Major Model Releases | Model | Company | Key Details | |-------|---------|-------------| | **GLM-5.2** | Z.ai (Zhipu) | 744B params, **1M token context window**, reasoning controls for long-horizon coding. MIT-licensed open weights. (June 17) | | **MAI-Code-1-Flash** | Microsoft | Microsoft's **first dedicated coding model**, converts written descriptions into source code. (Announced at Build, June 2) | | **MAI-Thinking-1** | Microsoft | Reasoning-focused model, private preview via Microsoft Foundry. | | **Gemini 3.5 Flash** | Google | Now the default model for Google Search's AI Mode, agents, and coding tasks. | | **Kimi-K2.7-Code** | Moonshot AI | Open-source coding model, ~30% fewer parameters than prior versions with improved coding performance. (June 10) | | **Nemotron 3 Ultra** | NVIDIA | 550B params / 55B active, open-weights. Announced at Computex. | | **Grok Imagine Video 1.5** | xAI | Launched June 16, now **#1 on Image-to-Video Arena** (+52 Elo over v1.0), outranking Sora 2, Veo 3.1, and Kling. | --- ### 2. 🛠️ Top New AI Tools & Products **From Product Hunt (June 15–21, top launches):** - **Upstream** (#1) — "The inbox designed for humans *and* agents" — 244 upvotes. A new category of shared AI-human workspace. - **Goldfish** (#2) — Press Option, it knows your work and replies like you — 218 upvotes. - **Framer 3.0** — Agents on the canvas, branching, and a new creator community. - **Claude Code Artifacts** — Live, shareable coding previews directly from Claude Code sessions. - **MakersClaw** — AI employees that live in Slack/Teams/Telegram. - **Swytchcode CLI** — Reliable agent access to 2,000+ APIs with durable state. - **AutoEdit** — Claude AI Video Editor for Premiere Pro. **From Product Hunt (June 8–14, top launches):** - **Bond** — "The AI to-do list that does itself" — #1 with 185 upvotes. - **Publora** — Publishing API for agents across 10 social platforms. - **Browse.sh** — "Give your agents muscle memory for automating the web." - **ElevenAgents by ElevenLabs** — Scale conversations without scaling teams. **From FutureTools.io:** - **Perplexity Brain** — A self-improving memory system for AI agents that tracks what worked/failed and improves overnight. +25% answer correctness, +16% recall. - **OpenAI Record & Replay** — Turns Codex workflows into reusable, shareable skills. - **Liquid AI LFM2.5 Retrievers** — Fast multilingual search models. - **MIT DAAAM** — Gives robots long-term memory in real-time via 4D scene graphs. --- ### 3. 📊 Enterprise & Platform Shifts - **OpenAI** added spend controls and usage analytics for ChatGPT Enterprise; upgraded health responses with GPT-5.5 Instant (factuality dropped 71% in two months). - **Anthropic** brought live artifacts to Claude Code and signaled Mythos/Fable 5 access could return "within days." - **ChatGPT's market share slipped below 50%** for the first time — a notable milestone. - **AlphaFold lead John Jumper left Google DeepMind for Anthropic** — a major talent move. --- ### 4. 🎯 Why It Matters **The "agentic" shift is no longer theoretical — it's productized.** - Products like Upstream, Browse.sh, Publora, and MakersClaw are built *for AI agents as first-class users*, not just for humans. The inbox, the browser, the social media API — they're all being redesigned around multi-agent workflows. **Coding models are proliferating across labs.** - Microsoft (MAI-Code-1-Flash), Z.ai (GLM-5.2), Moonshot AI (Kimi-K2.7-Code), Cohere (North Mini Code), and Google (Gemini 3.5 Flash) all shipped coding-focused models this month. This is creating real competition and price pressure in the AI coding space. **Memory and context are becoming differentiators.** - Perplexity's Brain, GLM-5.2's 1M token window, and MIT's DAAAM for robots all point to the same thesis: *the agent that remembers outperforms the agent that forgets*. This is moving from research to production. **Open-weight models are catching up.** - GLM-5.2 (MIT license), Kimi-K2.7-Code, and Nemotron 3 Ultra are all open-weight releases from major labs — giving developers more options beyond the big-three APIs. --- ### 5. 📚 Strongest Sources | Source | Coverage | Link | |--------|----------|------| | **AI Flash Report** | Today's model releases with verified sources | [aiflashreport.com](https://aiflashreport.com/topics/new-ai-model-releases.html) | | **FutureTools.io** | Curated daily AI news (June 14–20) | [futuretools.io/news](https://futuretools.io/news) | | **Product Hunt Weekly** | Community-voted best launches | [producthunt.com/leaderboard/weekly/2026/25](https://www.producthunt.com/leaderboard/weekly/2026/25) | | **Launch AI Jam** | Tracked 30 new tool launches this week | [launchaijam.com/new-ai-tools](https://launchaijam.com/new-ai-tools) | | **AIMADETools** | June 2026 model roundup | [aimadetools.com](https://www.aimadetools.com/blog/best-ai-models-june-2026-roundup/) | | **Kingy.ai AI Launch Radar** | Practical tool/agent/model roundup | [kingy.ai](https://kingy.ai/news/ai-launch-radar-june-8-2026/) | | **CNBC** | Microsoft MAI models announcement | [cnbc.com](https://www.cnbc.com/2026/06/02/microsoft-unveils-new-ai-models-lessen-reliance-on-openai-lower-costs.html) | --- **Bottom line:** This week marked the moment AI tools stopped being "assistant features bolted onto products" and became *the architecture* these products are built around — agents as users, memory as infrastructure, and coding models as a battleground across five+ labs simultaneously.