Welcome, humans.
Claude Opus 5.5 is absolutely wild. We planned a clean head-to-head benchmark between the two. Then Opus 5.5 showed up looking less like “another frontier model” and more like the thing that makes your benchmark designer surrender to the exponential.
So naturally, we made the tests much more ridiculous.
We are live NOW, back for Round 2 of GPT-6 Sol vs. Claude Opus 5.5.
This time, the models don’t get to answer questions. They have to build things.
That means creating, debugging, using tools, making product decisions, adapting when requirements change, inspecting their own work, and continuing until something actually works.
These are not multiple-choice benchmarks
Interactive Black Hole Lab: Build a scientifically useful black hole simulator with gravitational lensing, photon trajectories, controls, and explanations.
The Last Observatory: Use Blender to create and art-direct an entire miniature planet, observatory, astronaut, black hole, lighting setup, and animation.
AAA CAT DOOM: Push our increasingly questionable Cat Doom benchmark toward DOOM Eternal territory, with a much higher bar for gameplay, visuals, systems, and polish.
The Dark Souls Benchmark: Build a demanding game where mechanics, difficulty, atmosphere, level design, and actual playability all have to come together.
The progression: Can it build? → Can it create? → Can it invent? → Can it fix? → Can it adapt? → Can it ship something that feels like a real game?
Every model gets the same core instruction:
Do not explain how I could build this. Build it. Use the tools available to you, inspect your own result, and keep working until you believe it is finished.
We’ll compare GPT-6 Sol and Claude Opus 5.5 on completion, visual quality, judgment, autonomy, usefulness, and the extremely scientific category of “did it do something that made us yell?”
Opus 5.5 has the potential to make this round completely ridiculous. Come watch us find out whether GPT-6 Sol can keep up, and whether the benchmark charts survive contact with Cat Doom.
While you wait: catch up on this week’s chaos
If today is Round 2, these are the three episodes that got us here.
We tested GPT-6 Sol, GPT-6 Luna, and Claude Opus 5.5 on coding, reasoning, writing, agents, pricing, and everyday work. Opus 5.5 is the reason today’s benchmark got much harder.
Vincent walked us through OpenClaw’s jump from personal assistant to agentic computing platform: local and cloud workers, interactive widgets, persistent agents, automations, memory, Swarm, and the security controls that keep all of that from becoming chaos.
CoreWeave EVP Chen Goldberg explains why long-running agents change the infrastructure problem underneath AI: reliability, latency, security, orchestration, storage, networking, cooling, and power all have to behave like one enormous computer.
Yes, this has been a very normal three days.
Want the official model details? GPT-6 Sol + Luna | Claude Opus 5.5
Stay curious,
The Neuron Team
What do you want to learn about AI?Pick your favorite, then share any others in the "additional feedback" |
|
P.P.S: Love the newsletter, but don’t want these podcast and livestream announcement emails? Don’t unsubscribe.
