|
|
|
| // WEEKLY 04 · JUL 01 2026 |
|
OpenAI shipped its best model yet this week, and the government decided almost no one can use it. Anthropic's most powerful Claude just came back online. And there is a prompt below for the email you keep putting off.
|
|
| // SECTION 01 · NEWS |
|
|
The government decided who gets OpenAI's newest model
On June 26 OpenAI previewed GPT-5.6, a family of three models called Sol, Terra, and Luna, with sharp gains in coding, biology, and cybersecurity. Then it did something new: it limited the preview to a small group of vetted partners at the request of the Trump administration, which cited safety. OpenAI made its discomfort public, writing that it does not believe "this kind of government access process should become the long-term default" because it keeps the best tools from the developers, enterprises, and cyber defenders who need them.
Why it matters: for the second time in a month, the thing standing between you and the most capable AI is not price or performance, it is a government sign-off. First the Commerce Department pulled Claude Fable 5 from foreign access, now Washington is deciding which companies touch OpenAI's newest model first. Access is quietly turning into a policy question, not just a product one. Source.
|
|
› Fable 5 is back. On June 30 the US Commerce Department lifted the export controls it had placed on Claude Fable 5 and Mythos 5, and as of today, July 1, Fable 5 is live again across Claude.ai, Claude Code, and Cowork. Commerce Secretary Howard Lutnick signed off after a two-week review with Anthropic. Source.
|
|
› GPT-5.6 comes in three sizes. Sol is the flagship at 5 dollars in and 30 out per million tokens. Terra matches last generation's quality at half the cost, 2.50 and 15. Luna is the cheap, fast one, 1 and 6. If it ever reaches you, Terra is the one most people will actually run. Source.
|
|
› Anthropic pointed Claude at the lab. On June 30 it released Claude Science, a version of Claude wired into more than 60 scientific databases and research tools, aimed at drug discovery and pharma work. Source.
|
|
› Google's Gemini 3.5 Pro slipped to July. The June window Google promised on stage at I/O closed with no public release. It now points to July, citing more quality work after enterprise testing. Source.
|
|
|
| // SECTION 02 · THE TIP |
|
|
Most people accept the AI's first answer. The better move is to make it grade itself. After it gives you something, reply with "score that out of 10 for [what you care about], tell me what is holding it back, then give me the 10." Models are far better at spotting weak spots when you ask them to critique than at nailing everything on the first try, and the second version is almost always sharper. You are basically getting a free editor in one extra line.
|
|
Try it tonight: take the last thing AI wrote for you, paste it back, and say "score this out of 10 and give me the 10."
|
|
| // SECTION 03 · THE PROMPT |
|
|
For the message you keep putting off, the one where you have to say no, ask for money, or own a mistake. It gets the tone right so you actually hit send.
|
|
|
|
You are my email coach. I need to send a message I have been avoiding. Do three things:
1. ASK: if you are missing anything (who it is to, our history, what outcome I want, how blunt I can be), ask me before you write.
2. DRAFT: write it short, clear, and human. Firm without being cold, kind without being soft.
3. OPTIONS: give me two versions, one warmer and one more direct, and tell me when to use each.
Keep it to something a real person would actually send. No corporate filler.
Here is the situation:
[WHO IT IS TO, WHAT HAPPENED, WHAT I WANT]_
|
|
|
Try it tonight: pick the email sitting in your drafts and let it write the first version.
|
|
| // SECTION 04 · DECODE |
|
|
The newest, most capable AI models, the ones at the front of what the technology can do. They cost the most to train, they can do the most, and lately they are the ones governments watch closest. The line matters because new rules and restrictions almost always target frontier models, not the smaller, cheaper ones running underneath them.
|
|
Example: GPT-5.6, Claude Fable 5, and Gemini 3.5 Pro are all frontier models. That is exactly why this week's news was about who is allowed to use them, not how much they cost.
|
|
| Two more places I post the stuff that does not fit here. |
|
|
|
| ● STATUS · ONLINE · OPUSJAKE |
© OPUSJAKE // CRT V1 |
|
You're reading the web version of OpusJake Weekly. Subscribe to get the next one in your inbox.
All issues · OpusJake, 6620 W Fort St, Detroit, MI 48209
|
|