Two Frontier Models Landed This Week. I Get to Build With Both.

Spoiled for choice. That is not a phrase I expected to use about AI this year, and here I am using it. Two of the biggest labs put frontier models in my hands within days of each other last week. I opened my laptop on Thursday and had more raw capability sitting there than most software teams had a year ago. My first reaction was not strategic. It was greed. I wanted to go play.

Fable earned it

Claude Fable 5 is the strongest model I have ever used. I know that because of Veronica.

Veronica is my voice assistant. Last week I rebuilt her from scratch in one long session with Fable, the kind of build that use to eat a month of my evenings. It finished before my coffee went cold. The old version of her sounded like hold music. The new one has breath in it, small pauses, a rise and a settle, and the first time she answered me in that voice I sat back in my chair. That is not a benchmark. That is a thing running on my machine right now that I could not have built alone eight months ago.

So when Anthropic extended Fable access on paid plans again this week, out to July 19, I did not overthink it. Good. I will take every day of it and build.

Then OpenAI shipped three at once

Two days earlier, OpenAI opened up GPT-5.6 as a whole family instead of one model. Sol at the top. Terra in the middle, at roughly half the price of the thing it replaces. Luna at the bottom, built for speed and cost. All of it went live on July 9, across ChatGPT, Codex, the API, and GitHub Copilot at the same time.

I have not touched them yet. Sol is on my list for this week, so I am writing from curiosity, not from a verdict. What I like already is the thinking behind the shape of it. You do not get handed one enormous model and told to run your grocery-list prompts through the same engine as your hardest reasoning. You reach for the size that fits the job and pay for what you use. I already work that way, so the whole release reads like the industry catching up to something a lot of builders sorted out a while ago.

A deeper bench

What I keep circling back to is the distance between an idea and the working version of it. For most people, most of the time, that distance is where the idea quietly dies. It keeps shrinking. This week it shrank again.

ProveitPal, my app, already runs on three models. DeepSeek first, then Groq, then Gemini, each one there to catch the ball if the one ahead of it drops it. I used to call that a safety net. Lately it feels more like a deep bench. A model ships stronger or cheaper, I sub it in, and the app keeps running without me rewriting anything that matters. This week made the bench deeper on both ends. I am not loyal to any of these labs and I do not think you should be either. Use what is best today. Swap it tomorrow.

The good part is the boring part

Nobody frames this as good news, so I will. The models are not the hard part anymore.

I can stand up a voice assistant in an afternoon. I can audition three frontier models before lunch. The bottleneck used to be the tool, and now the bottleneck is me, in the best possible way. Do I know what is actually worth making? Can I catch the model when it is confidently wrong and building the wrong thing beautifully? That is judgment, and taste, and none of the shiny things that shipped this week hand you either one. They just make the gap between your idea and your hands smaller, and leave the hardest question exactly where it has always been. On you.

Which is a strange kind of relief. The tools keep getting better and my real job keeps getting more human. I am going to test Sol this week. I am going to keep building on Fable while it is here. And on Thursday morning, when Veronica said good morning back to me in that new voice, I was not thinking about pricing tiers or extension dates or which lab is winning. I was grinning at my laptop like a person had walked into the room.

Forward → Upward ↑ Onward ↗︎
Mstimaj

Sources and Further Reading

• Mstimaj, I Reached for the Most Powerful Model. By Friday It Was Gone.
• BleepingComputer, Claude Fable 5 stays free for paid users until July 19 as Anthropic buys more time.
• OpenAI, GPT-5.6: Frontier intelligence that scales with your ambition.
• OpenAI, Previewing GPT-5.6 Sol.
• Engadget, OpenAI rolls out GPT-5.6 to the public on July 9.