June 2026.4
Hey! This week had a strange mix: Open-weights models are catching up, image-generation companies are moving into medical hardware, and the old question of who controls the data behind the product is becoming harder to ignore.
📖 Story 1: GLM-5.2 is the new leading open weights model on the Artificial Analysis Intelligence Index
artificialanalysis.ai · Read
Z.ai’s GLM-5.2 is now the top open-weights model on the Artificial Analysis Intelligence Index.
The model scores 51 on the new v4.1 index, ahead of MiniMax-M3, DeepSeek V4 Pro, and Kimi K2.6. It keeps the same size as GLM-5.1 — 744B total parameters, 40B active — but adds a much larger 1M token context window and noticeable gains on several reasoning and agentic benchmarks.
The interesting tradeoff is token usage. GLM-5.2 is strong, but it also spends a lot of reasoning tokens per task, which makes “best open model” less simple than just looking at the benchmark score.
💬 HN Discussion
The HN thread was positive, but not blindly so. A lot of the discussion focused on reasoning efficiency: GLM-5.2 may be very capable, but some users found it slow and token-hungry at max effort.
Several commenters compared it to Claude Opus-class models, while others pointed out that these huge open-weights models are still far from easy local deployment. The recurring theme was: open weights are catching up fast, but frontier-level open models are still mostly data-center models, not laptop models.
📖 Story 2: Midjourney goes from generating cat images to full-body ultrasound scans
theverge.com · Read
Midjourney, best known for AI image generation, is moving into medical imaging with Midjourney Medical.
The company is working on a full-body ultrasound scanner that would place a person in water and use underwater sensors to reconstruct a 3D body scan in about a minute. The first version sounds less like a hospital device and more like a health spa product: repeated scans, body-composition tracking, and personal health data over time.
There is real hardware behind it too: the prototype uses ultrasound modules from Butterfly Network.
💬 HN Discussion
The HN thread was mostly skeptical but technically interesting. Radiologists and medical-device people pointed out that ultrasound has hard physical limits: air, bone, depth, and resolution all matter. The big question is whether this can become medically useful, or whether it is mainly a futuristic wellness product with impressive branding.
📖 Story 3: Gaslighting Openness
lucumr.pocoo.org · Read
Armin Ronacher published a sharp reminder that users should care where their data goes, especially as AI tools become more personal and more deeply integrated into everyday software.
This is also where European regulation matters. GDPR and the AI Act are often framed as obstacles to innovation, but they protect consumers from paying twice: once with money, and again with personal data.
We've seen this pattern before. Pokémon Go players thought they were just playing a game, while the data helped build mapping datasets later useful for robots and autonomous systems.
💬 Community Moment
It's actually le gros chaton
Mistral’s cat-themed naming joke keeps going. This time, even Mistral co-founder Arthur Mensch joined in, continuing the French wordplay with “le gros chaton” — “the big kitten.”
🛠️ Projects Worth Checking Out
- GitHub - elder-plinius/CL4R1T4S: LEAKED SYSTEM PROMPTS
- OpenCode Go | Low cost coding models for everyone
- GitHub - StarTrail-org/PixelRAG: The beginning of scalable pixel-native search.
- GitHub - PrefectHQ/fastmcp: 🚀 The fast, Pythonic way to build MCP servers and clients.
- GitHub - google-antigravity/antigravity-cli
