In partnership with
Welcome back. Anthropic shipped Claude Opus 5.5 on Monday, the smaller model that now beats their own bigger one. Google pushed Gemini 3.8 Flash into general availability the same week, and a browser agent called Jev is quietly eating the classification work nobody wanted to do by hand.
Today: the Opus 5.5 launch and the catch under it, the frontier's new shape, what is trending, the toolbox, a ten-minute tutorial, and one prompt to test any model on your own work.
Some teams never seem to stop moving. They're on Attio, the agentic CRM.
Every customer signal is captured in one shared context layer, always current and compounding. Agents and workflows build pipeline, chase every buying signal, and move deals forward, an always-on revenue engine running alongside your team.
With Attio, you’ll get:
Leads automatically prioritised and routed to the right rep
Expansion and risk signals caught the moment they land
Follow-ups written in your voice, already there when you arrive
Teams like Parallel, Turbopuffer, and Wordsmith build on Attio. Are you one of them?
Today in AI
1. Opus 5.5 is cheaper and beats the bigger Claude: Anthropic's new smaller model runs about 40% cheaper than Opus 5 and 30% faster, and it outscores the larger Fable 5.1 on most coding and agentic work. Input is now $4 per million tokens, output $20, cache reads down 60%. read the launch notes here.
2. Where it leads: SWE-bench Pro at 89.9%, top of the table, and Terminal-Bench 4.0 at 66.4%, ahead of GPT-6 Astra. It trails Astra on a couple of agentic research tests, both close enough to be noise.
3. The catch: independent testing by Vals AI found it gained big on code but dropped 4 to 14 points on legal, medical and tax tasks. A general upgrade is not an upgrade for every job.
What this means for you: move your coding work to Opus 5.5 today, and test before you switch anything legal, medical or tax.
From God of Prompt
You do not need the newest model to get a result tonight. The free prompt library is sorted by the job you are doing, so you pick the task and paste. Built and tested, filed by outcome.
From the frontier
The frontier stopped being about the biggest model.
The case for cheap and small. Opus 5.5 beats a larger model at a fraction of the cost, and Anthropic cut prices to match. When the smaller model wins, size stops being the pitch and price does the talking.
The case against calling it settled. The same model regressed on legal, medical and tax work. Cheaper and smaller is winning at code, not everywhere, and the domains it lost are the ones businesses pay the most to get right.
Where this goes. The labs are now competing on cost per useful task, not raw capability. That is good for you and bad for anyone selling the model as the product. What would prove me wrong: one model that tops the code benchmarks and the domain ones at the same low price. Nobody has shipped it yet.
In the know
The price war: developers doing the Opus 5.5 math out loud, 40% cheaper for better coding. [engagement number]
The regression: the Vals AI finding that a newer model got worse at legal and medical is the debate under the launch. [engagement number]
The classifier: Jev, the model that scores instead of writes, is still spreading through developer feeds. [engagement number]
Gemini Flash: Google moved 3.8 Flash into general availability the same week. [engagement number]
The gap: benchmark score versus real domain work, the thread tying the whole week together. [engagement number]
From God of Prompt
The AI Starter Pack is $20, one payment. Every guide for every major model, plus the cheatsheets you keep open while you work. Built for the person with ChatGPT open and no idea what to type.
Toolbox
Slot for your tool: reply to this email and we will feature your project in the next issue!
Custom Prompt Generator: describe the task, get a prompt written for it, then paste it into whichever model won your test. Built by me, lives on the site.
Tutorial
How to check if Opus 5.5 is actually better for your work
Pick one task you do often and already know the right answer to.
Open Claude and set the model to Opus 5.
Run the task and save the output.
Switch to Opus 5.5 and run the same task.
Paste both into a fresh chat with the prompt below and let it score them.
Keep whichever wins on your task, not the leaderboard.
Compare these two outputs for the task [task].
The correct answer looks like [answer].
Score each output from 0 to 100 and list exactly what each one got wrong.Prompt station
The model router
When you are not sure which model to use, let it decide for you. This is the opening, the full one lives in the library.
You are my model router.
Task: [paste what you are trying to do].
Given ChatGPT, Claude and Gemini, tell me which one to use for this and why, in three lines.
Then give me the exact prompt to paste into it...👉 open the full prompt. Built for ChatGPT, Claude and Gemini.
For coding, Opus 5.5 is the one to switch to today. For everything else, test first.
🔑 Robert from God of Prompt
P.S. Every prompt in the free library runs in Opus 5.5 with no changes. Same prompts, cheaper model.
