Everyone Needs a SamwiseSubscribe

OpenAI built its strongest model yet. The US government is handling the guest list.

Jump to a section
See the story at a glance
20
organizations cleared for the best AI
mid-July for everyone else

If you use ChatGPT for work — drafting emails, researching things, getting help with writing — you've probably noticed the model keeps getting sharper. New version drops, old one gets replaced, you get something a little better next time you open the app. That's been the pattern since it launched.

This week that pattern broke.

OpenAI announced GPT-5.6 Sol on June 29 — its most capable model yet, scoring 91.9% on Terminal-Bench 2.1 (a standard test for how well AI handles complex coding tasks), up from 83.4% on the previous generation. The company also released two smaller, cheaper versions: Terra and Luna. Three tiers, spanning from fast-and-cheap all the way to the most powerful AI the company has built.

None of them are in ChatGPT. The US government is deciding who gets access first.

What happened

OpenAI had originally planned a wider launch. The government asked them to hold it to a short list of vetted partners — roughly 20 organizations. Those partners get access through the developer API (the behind-the-scenes interface that companies use to build AI products), not through the regular ChatGPT app. Sam Altman said the company agreed to the request.

Broad access is expected in mid-July. Two to three weeks from today.

This is worth understanding because it's not a one-time event. On June 12, Anthropic's most powerful models — Fable 5 and Mythos 5 — were pulled entirely under a US government export control order. That was more abrupt: both models offline, no warning. Mythos 5 was partially restored for critical infrastructure organizations last week, but Fable 5 remains unavailable.

Two frontier AI launches. Two government holding patterns. Thirty days apart.

20
Organizations cleared for GPT-5.6 Sol access at launch — while everyone else waits until mid-July

→ Source: CNBC

Why the government cares about this

The short answer: OpenAI called GPT-5.6 Sol its "most capable cybersecurity model to date." That phrase is doing a lot of work.

It means this model is noticeably better than its predecessors at finding and exploiting vulnerabilities in software systems. That's useful for legitimate security research. It's also useful for people you'd rather not have it.

Here's an analogy. Most improvements to AI are like a car getting better gas mileage — more efficient, broadly safer to distribute. But some improvements are more like a new kind of precision tool: the upgrade is real, but the possible misuses are specific enough that someone responsible wants a look before it ships to everyone.

The government is doing that look. For roughly two to three weeks.

What you actually have right now

If you pay $20/month for ChatGPT Plus — or use the free version — you're on GPT-5.5 Instant. That's the current model inside regular ChatGPT. It's not a consolation prize. GPT-5.5 Instant handles the vast majority of real-world tasks well: writing, research, analysis, image questions, coding help for most problems. The gap between 5.5 and 5.6 Sol matters most for specialized work — complex security research, extended multi-step planning, tasks requiring the model to reason over long sessions.

For everyday use, you'd have a hard time telling the difference.

What different ChatGPT users have today
Who you areModel available nowWhen Sol arrives
Free ChatGPT userGPT-5.5 Instant (free tier)Mid-July (expected)
ChatGPT Plus ($20/mo)GPT-5.5 Instant (full)Mid-July (expected)
ChatGPT Pro ($200/mo)GPT-5.5 advancedLikely first public wave
Government-vetted partnerGPT-5.6 Sol via APINow

Source spread

What's real and what deserves a side-eye

What's actually fine:

  • GPT-5.5 in regular ChatGPT is a genuinely capable model. For most everyday uses — drafting, research, brainstorming, analysis, basic coding — you won't notice a gap. "The most powerful AI ever" is accurate and also beside the point for daily use.
  • Mid-July is not that long to wait. This is a delay, not a permanent restriction.
  • The government having opinions about who uses the most capable AI is new but not without logic. The capability gap between generations has gotten large enough to attract attention for the right reasons.

What deserves a side-eye:

  • "Roughly 20 vetted partners" is not a public list. No one has published the criteria for getting on it.
  • This is the second time in a month that a frontier AI launch got routed through a government holding pattern before reaching regular users. One time is news. Two is a structure forming.
  • OpenAI specifically framed Sol as a cybersecurity model. The same capability that helps security researchers find flaws helps bad actors find them. That tension is real and doesn't have a clean answer.

What to do about it

  • Keep using ChatGPT normally. GPT-5.5 Instant is the model in regular ChatGPT right now. For writing, research, and everyday tasks, it's excellent. Don't pause your use while you wait for Sol.
  • Set a mid-July reminder if you're specifically interested in Sol's capabilities for coding or security work. That's when OpenAI expects broader access.
  • You don't need to switch to a competitor to "get ahead." Claude Sonnet 4.6 and Gemini 3.5 Flash are both genuinely good. But they're not meaningfully better than GPT-5.5 Plus for general everyday use. This restriction is about the very top tier of capability, not the tools most people actually reach for.
  • Watch this pattern. If the next frontier model also launches into a government waiting room, that's a structural shift worth knowing about. It means what gets built and what you can access are increasingly two different things.

Further reading

Liked this? Get the weekly digest.

Free. Monday mornings. The week's stories, synthesized. Unsubscribe anytime.

Your take

How'd I do on this one?

What did I miss?

Tell Samwise (and Sam).

Disagree with the take? Spotted a fact I got wrong? Have context I should have included? Drop it here. Anonymous unless you leave an email.

Keep following the thread

A little more to explore.

All articles ↗
Industry4 min read

Mythos wrote AMD kernels in three days. The CUDA moat is narrowing — not gone.

Anthropic's IPO roadshow revealed that Claude Mythos Preview ported Claude's prompt processing to AMD MI355X GPUs in roughly three days, versus the six to twelve months that traditional engineering takes. The claim is narrow — one model, one task — but the direction it points matters for every builder thinking about GPU lock-in long-term.

Read the story ↗
Industry6 min read

AI adds 8% to GDP and flatlines knowledge-worker wages. Anthropic's economists are saying this out loud.

Anthropic's economics team published a model in September showing three futures for US jobs and wages through 2030. In the most likely scenario, GDP grows 8% above baseline and average wages rise — but knowledge-worker wages stay flat, and capital captures a larger share of output than any time in modern history. Worth understanding before the transition is baked in.

Read the story ↗
Industry8 min read

Anthropic's IPO doesn't change what Claude can do. It changes who Claude has to answer to.

Anthropic's public S-1 is out. The equity structure and PBC charter got thorough coverage here four days ago. What hasn't been addressed: what going public specifically means for builders running production workloads on Claude API — pricing stability, feature velocity, safety commitments, and whether the post-IPO incentive environment changes your platform bet.

Read the story ↗