On ARC AGI 2, Opus is just a smarter model and I think we see such big climb not because it is so good at thinking, but rather without thinking it is nearly impossible for it to do these puzzles well (same for me). E.g. GPT-5-Pro is below Opus 4.5 thinking 16k, but there's no way
@petergostev
-
Open Source Models Catching Up to Opus Performance
By
–
Ok if open source models can do in 6-8 months what Opus can do now, I'll believe that they are actually catching up
-
May I Meet You: Opus 4.5 Digital Canvas Art
By
–
'May I meet you?' by Opus 4.5 SVG on a digital canvas (2025)
-
Claude Opus 4.5 outperforms GPT-5.1-Pro in efficiency
By
–
Claude Opus 4.5 – multi-turn (8 edits). A really nice result, I wouldn't say it is beating GPT-5.1-Pro, but it didn't take anywhere near as much time. I haven't shared Gemini 3 Pro properly because I could never get anything that I liked, so Opus is definitely winning here! https://t.co/dCFOyDu0Sx pic.twitter.com/LRLt2FNk9a
— Peter Gostev (@petergostev) 24 novembre 2025Claude Opus 4.5 – multi-turn (8 edits). A really nice result, I wouldn't say it is beating GPT-5.1-Pro, but it didn't take anywhere near as much time. I haven't shared Gemini 3 Pro properly because I could never get anything that I liked, so Opus is definitely winning here!
-
Codex CLI Pull Requests Growth Compared to Gemini CLI
By
–
Counting PRs is a little silly, but there's a huge shift in the Codex CLI PRs since August – though still behind Gemini CLI PRs, which I did not expect. New Codex model would be pretty cool to have.
-
16,800 Academic Papers Reference GPT-4 in 2025
By
–
There are 16,800 papers on Google Scholar that were published in 2025, which reference GPT-4, but not any future versions of GPT, e.g. no GPT-4o, 4.1, 4.5, or 5.
-
ChatGPT to Continue Shaping the Legal Profession
By
–
I'm glad ChatGPT will carry on the legal profession
-
ChatGPT Full Assistant System Prompt Revealed
By
–
My ChatGPT Pulse this morning titled "Full assistant system prompt"
-
Cerebras and GB200 Dominance in AI Hardware Strategy
By
–
Pretty much, zai were hinting heavily, one of the cognition engineers talked about thousands of GB200, and cerebras are marketing it heavily
-
SWE-1.5 Model Architecture: Base Model, RL Training, Inference
By
–
How @cognition /
@windsurf new SWE-1.5 model was built (probably), based on piecing together bits of information that was shared. Base model by: @Zai_org possibly GLM-4.6
RL Training: @nvidia on 'thousands' GB200 NVL72
Inference: @cerebras at 950 tks/sec