All three went GA on July 9 at Sol $5/$30, Terra $2.50/$15, and Luna $1/$6 per million input/output tokens, each with a 1M-token context and 128K output (per TechCrunch). Sol is now the ChatGPT default. Sam Altman told CNBC that Sol is 54 percent more token-efficient on coding than prior OpenAI models, a vendor claim TechCrunch flagged as not independently verified. The benchmarks split: Sol leads the Artificial Analysis Coding Agent Index 80 to Fable 5's 77.2, but trails on SWE-Bench Pro at 64.6 percent against Fable 5's 80 percent (per Simon Willison).

Why it matters. Sol takes the headlines, but Terra and Luna are where the cost math works. Reach for Sol only if the coding-index edge earns back the top-tier price; otherwise Luna at $1/$6 (per TechCrunch) handles most jobs. Read our take.

Shipped July 8 with full-duplex audio, it routes the hard questions to GPT-5.5 in the background while keeping the conversation live. Read our take.

The three named firms can now apply to purchase NVIDIA H200s, with the total quota under 200,000 units (per Bloomberg and The Japan Times, citing The Information). US export law already cleared the H200 for China back in December 2025; this reverses Beijing's own domestic block. NVIDIA had reported zero H200 revenue from China.

Why it matters. DeepSeek's compute ceiling just rose, which could pull its V4 timeline forward and reshape the open-weight race. Read our take.

The new desktop app (Mac now, Windows rolling out) merges Chat, Work, and Codex, and existing Codex projects and settings carry over automatically. The standalone Atlas browser gets an August 9 end date, under a year after its October 2025 launch; its capabilities move into the desktop app's built-in browser and a Chrome extension (per 9to5Mac and TechCrunch).

Why it matters. One surface instead of three. If you built a workflow around standalone Codex or Atlas, plan the migration before August 9. Read our take.

Live July 9 for Free, Pro, and Max, it breaks down your AI habits across a four-dimension fluency framework. TechCrunch noted the dashboard also nudges you to keep using Claude. Read our take.

The acquisition changes the math. Our pick: Make for code teams, n8n for data engineers, plus the exact exit criteria if you decide to stay put. Read it. Try Make.

OpenClaw is now backed by an OpenAI, NVIDIA, and Microsoft 501(c)(3). Self-host it at $15 to $40 a month, or pay Lindy from $49.99; we map which profile each one fits. Read it. Try Lindy.

Fresh July pricing: Cloudways wins for managed n8n and MCP server setups, DigitalOcean wins for Gradient ADK raw control. Per-team pick inside. Read it. Try Cloudways.

Vision, Kimi K2.7, browser tools, and cost centers all shipped, pushing our rating to 4.4 with a plan pick per dev persona. Read it.

Grok 4.5 hit Cursor at $2/$6 per million tokens on July 8. The guide stacks it against Sol and Fable 5 on published benchmarks and names the model each persona should run now. Read it.