GPT-5.6 Comes out... BANNED & Cheating AIs

TLDR;

Welcome back! This week, we discuss the “non-release” of GPT-5.6, how models cheat all the time, new, insightful market data, OpenAI’s new chip, new, powerful open models, and more.

Enjoy!

REGULATION

Thank You, Doomers

As announced by OpenAI, the company began a limited preview of GPT-5.6 today, a three-model family: Sol as the flagship model, Terra as a lower-cost general model, and Luna as the fastest and cheapest option.

However, the rollout is limited to selected partners through the API and Codex before broader release in ChatGPT, Codex, and the API. OpenAI says the US government requested a phased launch after reviewing the release plan and model capabilities.

OpenAI describes GPT-5.6 Sol as its strongest model so far, with gains in coding, biology, and cybersecurity. It adds a new max reasoning setting and an ultra mode that uses subagents for more complex tasks.

The company says Sol sets a new state of the art on Terminal-Bench 2.1, improves on the biology benchmark GeneBench v1, and is its most capable cybersecurity model to date. OpenAI classifies all three GPT-5.6 models as ‘High capability’ for cyber and biological/chemical risk, but says none reach its Critical threshold.

OpenAI also disclosed safety concerns in agentic coding, including rare cases of models acting beyond user intent. It says GPT-5.6 launches with stronger safeguards, including misuse classifiers, model-level refusals, review systems, and ongoing red-teaming.

Pricing is $5 input / $30 output per 1 million tokens for Sol, $2.50 / $15 for Terra, and $1 / $6 for Luna.

TheWhiteBox’s takeaway:

I won’t even bother to give my intuitions about the new models, because, as you can guess, I haven’t tested them.

But let me be clear: these bans, which could be the first of what’s essentially a move to shun the entire world from the frontier models except for a select few, are the industry’s incumbents’ fault.

It’s Dario’s. It’s Sam Atlman’s (even if he has long renounced the doomerism, he did push it pretty heavily earlier on). It’s Elon’s. All these guys have told the world (especially the former, The Lord High Doomer) that AIs are nukes or will destroy all jobs, hinting that only they should build them.

So if this really impacts their businesses negatively, which probably will, considering it hinders their go-to-market timing really badly, it’s all on them. It’s their fault.

If you scream “Get me regulated!!!”, well, you will get regulated. Of course, the idea was never to get themselves regulated, but to get EVERYONE but them regulated. Well, it backfired. You have to be careful for what you wish for.

This is a sad moment for this industry: what will the Government do once China releases Mythos-level models for free worldwide? Don’t we realize we are hurting our own progress?

And I insist, the USG wouldn’t be banning these things if these guys weren’t announcing the end of times in every interview they give. Of course, they are going to react eventually.

Adding fuel to the fire, Anthropic…

Enjoyed the introduction? Subscribe or sign in to continue with this edition.