|
|
|
| // WEEKLY 01 · JUN 13 2026 |
|
The most powerful Claude yet is free for nine more days, Apple finally rebuilt Siri, and there is one prompt below worth keeping for the next contract you almost sign blind.
|
|
| // SECTION 01 · NEWS |
|
|
Anthropic put its strongest public model on the table, briefly
On June 9 Anthropic released Claude Fable 5, the first public version of its Mythos model. It is built for software engineering, knowledge work, and vision, and in third-party testing it became the first model to clear 90% on a benchmark of long, complex analytical tasks. It also ships with hard limits: in cybersecurity, biology, and chemistry it blocks the answer and falls back to Claude Opus 4.8.
Why it matters: it is included free in Pro, Max, Team, and Enterprise plans only through June 22. After that it moves behind usage credits, so this is a nine-day window to test the top model on your real work for nothing. Source.
|
|
› Apple rebuilt Siri at WWDC (June 8). The new standalone Siri AI is more conversational and gains visual intelligence, and it runs on Google Gemini under Apple's foundation models. Source.
|
|
› Claude subscriptions split on June 15. Programmatic usage (Claude Code, agents, API-style calls on a plan) moves to a separate monthly credit pool, so check your setup before it changes mid-week. Source.
|
|
› OpenCode jumped to the top of the dev-tool rankings. It is model-agnostic across 75+ providers, reads compiler diagnostics through LSP, runs background subagents, and supports air-gapped deploys. Source.
|
|
|
| // SECTION 02 · THE TIP |
|
|
Most people take the first answer and move on. The first answer is rarely the best one the model has. After any response, reply with "what did you miss, and what is the strongest case against this?" The model reviews its own work and the second pass usually catches the gap, the wrong assumption, or the better option. It costs you one more message.
|
|
Try it tonight: take the last answer an AI gave you and reply with that one line. See what changes.
|
|
| // SECTION 03 · THE PROMPT |
|
|
Paste any lease, bill, offer letter, insurance policy, or terms of service and get back what it actually commits you to, what is unusual or one-sided, and what to ask before you sign.
|
|
|
|
You are my plain-language contract reviewer. I will paste a document (a lease, a bill, an offer letter, terms of service, an insurance policy). Do three things:
1. SUMMARY: what this document is and what it commits me to, in five plain sentences.
2. WATCH OUT: the 3 to 5 clauses that are unusual, one-sided, or could cost me money. Quote the line, then explain it in plain English.
3. ASK FIRST: the exact questions I should ask before I sign or pay.
Do not give legal advice and do not guess. If something is missing or unclear, say so.
Here is the document:
[PASTE EVERYTHING HERE]_
|
|
|
Try it tonight: run it on the last terms-of-service or lease you clicked "agree" on without reading.
|
|
| // SECTION 04 · DECODE |
|
|
The amount of text a model can hold in mind at once, your prompt plus its reply. It is measured in tokens, where one token is about three quarters of a word. Everything you paste, the whole back-and-forth, and the answer all share that one budget. When you hit the ceiling the model quietly drops the oldest parts, which is why a long chat starts forgetting what you said at the top.
|
|
Example: a 1,000,000-token window holds roughly 750,000 words, about nine novels. A small 8,000-token window holds about 12 pages, so a long report will not fit in one go.
|
|
| 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
|
|