🤚 The Open-Palm Model Launch

OpenAI has released GPT-5.6, a family of three models named after celestial bodies — because when your company is valued at several hundred billion dollars, you stop naming things after numbers and start naming them after the cosmos.

• Sol is the flagship — the workhorse, the one Sam Altman wants you to compare to Anthropic’s Fable 5
• Terra is the mid-tier model, for those who want intelligence but have a budget meeting at 3 p.m.
• Luna is the economy option, for people who think Opus 4.8 charges too much but still want to feel something

Sol scores 80 on the Artificial Analysis Coding Agent Index, which is 2.8 points above Fable 5 — while using less than half the output tokens, taking half the time, and costing approximately one-third less. Sam Altman himself noted that Sol achieves “54% more token efficiency when it comes to AI coding tasks,” which is the kind of specific metric that sounds devastating until you remember that benchmarks are the astrology of the AI industry.

The models are available across ChatGPT, Codex, and the OpenAI API as of July 9, along with a new product called ChatGPT Work for enterprise teams across desktop, web, and mobile.

The pricing is aggressive:

• Sol: $5 input / $30 output per million tokens
• Terra: $2.50 / $15
• Luna: $1 / $6

For context, Fable 5 costs roughly three times what Sol does. OpenAI has essentially walked into Anthropic’s living room, looked at the price tag on the couch, and asked “what if this were cheaper and faster?”

OpenAI also declared GPT-5.6 its “strongest cybersecurity model yet,” with particular capabilities in threat modeling, code review, patching, and blue teaming. Which is convenient timing, given everything else happening at the company this week.

👐 The Two-Handed Partnership Anxiety

But the real drama isn’t in the model weights — it’s in the phrase “preferred model.”

OpenAI has declared GPT-5.6 the “preferred model” for Microsoft Copilot 365, a designation that no one, including the publication that reported it, can quite define. What does “preferred” mean when Bloomberg has been reporting that Microsoft is quietly replacing OpenAI software with its own in-house models called MAI in Word and Excel to cut costs?

OpenAI’s official response was a masterclass in saying nothing with great conviction: “Our partnership with Microsoft has always been about bringing the benefits of advanced AI to more individuals and organizations, and we’re excited to continue building on that shared commitment.”

Translation: Please stop asking about the divorce. We’re still wearing the ring. The ring may be decorative at this point, but we are wearing it.

The “breakup chatter” refers to mounting evidence that Microsoft — which has invested approximately $13 billion in OpenAI — is hedging its bets. Building your own models while your partner launches new ones is the corporate equivalent of downloading a dating app while sitting next to your spouse at dinner. Everyone can see the screen.

🌿 The Gentle Awakening

Meanwhile, Fidji Simo — the woman who was literally OpenAI’s number two, CEO of Applications, with the COO, CFO, and CPO all reporting to her — has stepped down after just 14 months due to a relapse of a neuroimmune condition. She’s transitioning to a part-time advisory role.

Sam Altman posted on X that this “sucks,” which is both admirably honest and alarmingly casual for the departure of your most senior operational leader during an IPO runway.

Simo joins Kate Rouch (CMO) and Kevin Weil (CPO), both of whom left for health reasons in April. Brad Lightcap, the COO, has been shuffled into a “special projects” role — which, in corporate taxonomy, sits somewhere between “we still like you” and “your desk is now in the supply closet.”

That’s four senior executives reshuffled or departed in three months. OpenAI’s leadership page is starting to look like the departure board at Heathrow — lots of changes, frequent delays, and no one’s entirely sure which flights are still operational.

No replacement has been named. Speculation centers on Denise Dresser, who joined as Chief Revenue Officer in December after serving as Slack’s CEO and spending 14 years at Salesforce. She has the operational gravity. Whether she wants to inherit a org chart that’s been through a paper shredder is another question.

👑 The Gold-Leaf Copyright Problem

And then there’s the matter of the New York Times, which has accused OpenAI of concealing evidence in their copyright trial — the kind of allegation that tends to make judges irritable and opposing counsel giddy.

The allegations are specific and brutal:

• OpenAI claimed it couldn’t search its training corpus for copyrighted content — but data privacy engineer Vinnie Monaco apparently revealed in an April deposition that they’d already conducted exactly those searches internally
• OpenAI assembled approximately 78 million de-identified ChatGPT conversations to internally assess infringement levels — while telling the court it lacked the tools to do so
• A project called “Project Giraffe” deployed a Bloom filter tool that detected and recorded content regurgitation in outputs — deployed after the lawsuit was filed
• OpenAI allegedly deleted billions of ChatGPT outputs following the lawsuit filing and submitted chat log samples with redactions so heavy the court called them “unusable”

Lead counsel Ian B. Crosby summarized the situation with surgical precision: “If OpenAI genuinely believed that copying our clients’ journalism was fair and legal, it wouldn’t have hid the truth.”

OpenAI spokesperson Drew Pusateri responded: “We’ll continue defending our users’ privacy and the long-established principles of fair use.”

The Times is requesting sanctions, exclusion of the 20-million chat log sample as unreliable, and that the court accept unredacted logs would demonstrate substantial regurgitation. They also want OpenAI to pay their legal fees, which — given this is a case against a company pursuing a trillion-dollar IPO — may be the most reasonable demand on the list.

“The model beats Fable 5 on benchmarks, costs a third as much, and was announced on the same day the company lost its number two executive and got accused of destroying evidence — but sure, let’s talk about token efficiency.” — The Slap of Wisdom Benchmarks Desk, currently scoring its own internal stability at a confident 80 out of 100