Sol, Terra, and Luna shipped June 26 to roughly 20 approved partners only. Per OpenAI's announcement, pricing splits the field: Sol at $5 in / $30 out per million tokens, Terra at $2.50 / $15, Luna at $1 / $6, with Sol adding an "ultra" mode that routes hard tasks across sub-agents. Why it matters. The 5x price gap from Luna to Sol means you pick the tier per task, not per app; broad rollout is promised "in coming weeks," so plan the routing now.

Per Musk on X on June 27, the next xAI model release "will be called Grok 4" with a planned launch "just after July 4th." The team needs "one more big run" to complete a specialized coding variant. Why it matters. A coding-specialized Grok 4 arriving the week of July 4 will hit an already-crowded benchmark season alongside GPT-5.6's broader rollout.

Per Bloomberg, OpenAI may push its listing to 2027 with Altman calling any cut to the $1T valuation target a "nonstarter." SoftBank fell 13% in one session, erasing $38 billion in market cap, and still carries a $40 billion bridge loan due March 2027. Why it matters. The company underwriting much of OpenAI's compute just got more fragile; watch for a financing crunch that could reshape capacity and pricing downstream.

Mistral is in early talks for about EUR 3B ($3.5B) at a roughly EUR 20B ($23B valuation per Bloomberg via Yahoo Finance), near double its last mark. Alongside it, OCR 4 lands at $4 per 1,000 pages, 170 languages, an 85.20 OlmOCRBench score, and a 72% human-annotator preference rate, with a self-hosted container. Why it matters. If you process documents at volume, a self-hostable OCR at that price and language coverage is a real line item to re-bid against your current vendor.

Samsung Group presented South Korea President Lee Jae Myung with a plan to commit roughly 1,000 trillion won ($648B) over the next decade to chip manufacturing, AI data centers, batteries, and displays in South Korea. Up to 300 trillion won targets new semiconductor fabs. Why it matters. Samsung is signaling it intends to keep pace with TSMC's AI-infrastructure buildout domestically rather than cede the leading-edge fab race.

Patronus AI closed a $50M Series B (Greenfield Partners, $70M total) and launched Digital World Models: large-scale simulation environments where AI agents train via reinforcement learning, rewarded for task completion and penalized for errors. Revenue grew 15x+ in the past year; frontier AI labs and hyperscalers make up the majority of its customer base. Why it matters. Simulation-based agent evals are gaining fast on static benchmark sets; this is the infrastructure play behind that shift.

From that date the European Commission can fine general-purpose model providers up to EUR 15M or 3% of global annual revenue for breaching GPAI transparency and systemic-risk duties. OpenAI, Google, Anthropic, Meta, and Mistral all fall under it where their models are offered in the EU. Why it matters. If you ship a product into the EU on top of these models, your provider's compliance posture is now your exposure too; ask for their GPAI documentation before August.

Wire LLM steps into a workflow without code. For a team already living in a no-code stack, the new agent steps move Make from "nice glue" to a real automation layer worth the price. Try Make →

An MCP server lets an agent schedule and manage sends for you. Write-access is real but gated to the Scale plan, so price that in before you commit. Try Beehiiv →

For a team already living in a no-code stack, the new agent steps move Make from "nice glue" to a real automation layer worth the price. Solo builders on a tight budget can wait. Read the review →

For a daily-driver IDE user, yes; the agent mode still earns the seat. If you only touch code weekly, the free tier covers you. Read the review →

The write-access MCP is the real unlock for hands-off scheduling, but it is gated to the Scale plan, so price that in before you commit. Read the guide →

The three-step org-level install path so every engineer gets the same plugins on day one, instead of each person configuring their own. Read the guide →