# Closed labs are tightening up while open models flood in — 3 AI stories for builders (June 15)

> Source: <https://dev.to/danio_dev/closed-labs-are-tightening-up-while-open-models-flood-in-3-ai-stories-for-builders-june-15-12d2>
> Published: 2026-06-15 13:00:00+00:00

Three things crossed my feed today, and together they tell one story: the big closed labs are getting squeezed, while open, self-hostable options keep piling on. Here's the builder's-eye view.

A coalition of **42 state attorneys general** (led by New York's Letitia James) served OpenAI a subpoena. They want records on advertising, user retention, personal and health data, how ChatGPT treats minors and seniors, and the model's tendency toward **sycophancy** — just agreeing with the user.

It landed days after OpenAI's confidential IPO filing. For those of us shipping on top of ChatGPT, the signal is simple: expect tighter data and safety rules on consumer-facing AI, and more scrutiny of how your app handles vulnerable users.

Source: [CNBC](https://www.cnbc.com/2026/06/12/openai-says-its-engaging-constructively-with-state-ags-.html) · [The Next Web](https://thenextweb.com/news/openai-state-attorneys-general-investigation-ipo)

Anthropic is retiring **Claude Sonnet 4 and Opus 4** (Opus 4.1 is already deprecated and retires soon). If your app is pinned to those old model IDs, calls start failing — so repin to a current model now.

This follows Claude Code and the Agent SDK moving to a **separate paid credit pool**. If you build on Claude, both changes are worth a quick audit of your config and your billing.

Source: [Anthropic — Model deprecations](https://platform.claude.com/docs/en/about-claude/model-deprecations)

Six competitive open-weight releases shipped in about two weeks — Qwen 3.7, DeepSeek V4.1, Hunyuan Large 3, ERNIE 5.1, Doubao Pro and GLM-6. **DeepSeek V4.1 Flash** topped Hugging Face's trending list within a week.

While the closed labs tighten access, free and self-hostable options keep stacking up. If you've been renewing a closed plan on autopilot, it's a good moment to actually test an open model on your workload.

Source: [Hugging Face trending — Presenc AI](https://presenc.ai/research/huggingface-trending-models-june-2026)

I run a small daily AI-news channel and write these up as I go — still figuring out the workflow, so feedback welcome. If short daily recaps are useful, the video version is embedded above.