Good morning. OpenAI dominates the wire today with a triple-header — GPT-5.6, a reworked “ChatGPT Work” app, and the quiet exit of its number two executive — all landing the same morning the New York Times filed sanctions alleging OpenAI hid evidence in the copyright trial. Meta and Anthropic also showed up, but the story is really about how much OpenAI can push out in one news cycle, and how messily some of it lands.
GPT-5.6 arrives in three flavors, with a naming problem. OpenAI released GPT-5.6 as a family — Sol, Terra, and Luna — after a brief regulatory hold that limited access to government-approved orgs, per The Verge. Sol posted a new SOTA on ARC-AGI-3 (7.8%, the first frontier model to beat an ARC-AGI-3 game) and OpenAI claims 54% better token efficiency for coding, though TechCrunch notes the Fable 5 comparisons look suspiciously favorable. HN commenters were less impressed with Terra’s coding — one vibe-coded RTS test put it near GPT-5.5 and behind Sonnet 5 — and thoroughly baffled by the naming. “I really wish there was just an easy guide on when to use Sol vs Terra vs Luna,” one wrote, echoing a dozen others.
ChatGPT Work merges Codex into the main app, and users are lost. Alongside the model, OpenAI shipped ChatGPT Work, folding Codex into a unified desktop app with Slack, Gmail, and Drive connectors. The HN reaction was rough: casual chats got demoted to a small popup, the old app was renamed “ChatGPT Classic” (implying deprecation), and toggling between Work and Code modes seems to do little beyond tweaking verbosity. Anthropic pulled a similar Chat-vs-Cowork split yesterday, but commenters generally felt Anthropic’s execution was cleaner. “Leave what’s working alone. Build new adjacent things,” one wrote — the sort of feedback you get when you ship a launch and a reorg in the same week.
Fidji Simo steps back as OpenAI’s number two. OpenAI’s CEO of Applications and de facto second-in-command is moving to a part-time advisory role after a chronic neuroimmune condition flared up in April. The Verge notes she joins a run of recent C-suite exits — COO Brad Lightcap, CMO Kate Rouch, CPO Kevin Weil — leaving Altman thinner at the top just as IPO chatter picks up. Greg Brockman has absorbed product strategy and appears to be the person behind the “unified agentic platform” push visible in today’s launches.
The NYT says OpenAI hid evidence — and deleted billions of outputs. In a much less celebratory filing, the New York Times and Daily News allege OpenAI concealed a ~78 million de-identified conversation database and an internal regurgitation-detection tool called “Project Giraffe,” while telling the court such searches were technically infeasible. They also claim OpenAI deleted billions of ChatGPT outputs after the suit was filed, violated preservation orders, and handed over a heavily redacted, largely unusable sample. The plaintiffs want sanctions and want the 20-million-log defense sample thrown out.
Meta enters the coding race with Muse Spark 1.1. Meta launched Muse Spark 1.1 at $1.25/$4.25 per million tokens, pitched at bug fixing, code migrations, and multimodal workflows. It’s late to the party against Claude and GPT — Zuckerberg’s first X post in three years hyped more models coming — but the pricing is aggressive, and public preview credits should generate a lot of quick reviews this week.
Anthropic ships a public J-space demo. Following up on last week’s interpretability paper, Anthropic and Neuronpedia released an interactive tool for the “Jacobian lens” that surfaces Claude’s internal concept space. The finding worth flagging: what the model narrates about its reasoning doesn’t always match what J-space shows it’s actually computing — useful ammunition for anyone building on chain-of-thought as a safety signal.
That’s the day. If you’ve been putting off learning your Sols from your Terras, you have a weekend to figure it out — and OpenAI has a weekend to figure out where casual chat is supposed to live.