Nice

Never buy the first variant of something. Wait for the revision model

But Terra isn’t a coding model? Why not test it on computer use? What it was built for.

Sonnet has its cases so don’t opus and fable. Limiting yourself to a single model limits you just from my experiance

It was never meant to be an opus competitor

Stop using only opus and use the other models for use case specific things. Your mileage will last longer and have better outputs

If your workflow is more coding then general use, I’d say Claude is your play. If your a more well rounded person using it for different things like images ect codex seems to be the move. Personally I find instead of having Claude max20 I have both $100 from ChatGPT and Claude. Both suite me just well together

I mean technically it is, at what it’s built for. Computer usage.

No one said it bears opus 5.5 so why would it? It’s not even the same class.

Having access to the tools doesn’t mean you have the capability to do everything the same. Businesses have been this way since the dawn of time. Some do one thing really well and others do a lot decently. How many ai pop ups show up and then at the first sign of trouble fail? Simply having the use of ai doesn’t mean successful either. You still need to know what your doing. So just do your thing, stand out the best way you know how and be there for your customers and employees

Tell that to everyone else lol I agree with you there.

And the fact still stands. Terra was not meant to replace opus or be the next big code model. It has other uses which is where is t is better.

This is literally the use cases I see for these models…

I’d say Luna and haiku are more of the same class not Terra

GPT-6 Astra For the hardest problems: architecture, deep research, complex debugging, and critical reviews. The pick when quality and judgment matter most. GPT-6.1 Sol The coding workhorse. Great for building features, refactoring, writing tests, and handling substantial projects while keeping usage reasonable. GPT-6.1 Terra For visual reasoning, browser/app tasks, and clear writing. A good fit for screenshots, visual QA, polishing content, and small, well-defined code fixes. GPT-6 Luna For focused, repeatable tasks: summaries, extraction, classification, and simple edits. The efficient option when you don’t need heavy reasoning.

No one is claiming that Terra is an opus replacement. Nor should it be looked at in that way. Each model is best at something and this is typically the way I would use them. You don’t need to run Astra for everything. Your tokens could last a lot longer using them the right way. GPT-6 Astra For the hardest problems: architecture, deep research, complex debugging, and critical reviews. The pick when quality and judgment matter most. GPT-6.1 Sol The coding workhorse. Great for building features, refactoring, writing tests, and handling substantial projects while keeping usage reasonable. GPT-6.1 Terra For visual reasoning, browser/app tasks, and clear writing. A good fit for screenshots, visual QA, polishing content, and small, well-defined code fixes. GPT-6 Luna For focused, repeatable tasks: summaries, extraction, classification, and simple edits. The efficient option when you don’t need heavy reasoning.

Tested /clear vs one long Claude Code session on a 197-file repo: 3 unrelated tasks, 3 runs each way. Every run passed all hidden tests, and one session was 14% cheaper ($0.137 vs $0.159 a run), but its context grew to ~31K tokens vs ~21K.

Why are we comparing it to opus? It was never a replacement for opus

Nice find. Does the patch do anything for decode, or is it purely prefill? And did you see the same gain on a quantized model, or only BF16?

Do you know where Sonnet's extra cost per task came from, more turns or more context re-read each turn? Curious whether it's the model or how the harness drives it.