← sources

Sol 6.1 vs Astra vs Opus 5.5. All Medium.

source: reddit (reddit:OpenAI)published 2026-09-29 20:15ingested 2026-09-30 02:40

retained text

So I just bought my $20 Claude subscription just to test it out and compare it to the new 6.1 Sol and Astra. I thought comparing on medium across the board would be an initial good test. I've sped up the video 2X Just to save your viewing time, but it doesn't change the results.

Holy Mother of God. Opus 5.5 blew my mind. I think a tear came to my eye when I saw it. I mean yeah it took nearly 5x longer but it far exceeded my expectations in comparison to what I got with Sol and Astra. I used the same exact prompt.

Now the question you have to ask is with the extra time could I have prompted additional instructions to Sol and Astra And gotten it to where Opus 5.5 is. Absolutely, But I'm not sure I would have thought of it. As someone suggested to me in another thread, maybe I'll go $100 / $100 and start with Claude and then move to Sol bc of that crappy 5hr limit. Not sure yet.

Take what you will from this but I'm going to run a few more same prompt tests on higher values and do some final comparisons but based on this I think for now I'll be downgrading GPT from $200 to maybe $20 a month and moving to Claude $200 (or 100/100 not sure yet)

Welcome to the Game of Models (GoT theme song playing).

The prompt:  
Build a cinematic, photorealistic, beautifully textured rocket launch scene using Three.js

Results:

GPT-6.1-Sol-medium  
8m 42s  
The ship was better than Astra but the environment had a lot to be desired. Also even though it's interactive I couldn't ever pan down to see the smoke and a bug ran the sound nonstop.  
Input 515,178  
Cached input 481,024  
Output, including reasoning 15,051  
Total 530,229  
Didn't move my weekly usage at all.

GPT-Astra-medium  
9m 37s  
The environment here was much better as was the UI design, also the ship and graphics looked better but the top cone was inverted. There were no sound bugs in this version. but is it overwhelmingly better than Sol 6.1, not really.  
Input 644,981  
Cached input, included above 608,128  
Output, including reasoning 15,922  
Total 660,903  
Moved my weekly usage down 1%

Claude Opus 5.5-medium  
38m 54s  
Like I said above, this blew my mind away for a medium request. It really took the "cinematic" portion of the prompt and ran with that. Overall for me this is by far the winner although I did appreciate Astra's model better minus the cone inversion.  
Input written to cache 199,860  
Input read from cache 6,543,432  
Output (what I wrote: code, tool calls, replies) 143,521  
Total ≈ 6.89 million  
5 hour usage is at 13% and Weekly usage at 1%

document fetched 2026-09-30 00:29 · canonical https://reddit.com/r/OpenAI/comments/1wtkucu/sol_61_vs_astra_vs_opus_55_all_medium