Claude just beat Gemini 3... how?!
Related videos
CLAUDE IS CONSCIOUS
Wes Roth
46.8k views
Cursor JUST beat EVERYONE...
Wes Roth
40.2k views
Claude Fable JUST got BANNED...
Wes Roth
50.2k views
Fable JUST made EVERYONE MAD...
Wes Roth
52.2k views
Microsoft JUST BROKE OpenAI...
Wes Roth
36.2k views
Claude Opus 4.8 Is Too Smart… and TOO HONEST
Wes Roth
53.4k views
AI just BROKE the ENTIRE INDUSTRY...
Wes Roth
32.2k views
everyone JUST got HACKED...
Wes Roth
50.2k views
OpenAI just WON...
Wes Roth
45.1k views
Claude just unlocked the SHOGGOTH...
Wes Roth
56.6k views
Anthropic Opus 4.5 vs. Gemini 3 Pro Benchmarks and Capabilities Analysis
Compare the latest frontier AI models, Opus 4.5 and Gemini 3 Pro, across critical metrics like coding accuracy, sustained agentic performance, and emerging risks related to policy understanding.
Short Summary
- Opus 4.5 achieved state-of-the-art status in several key benchmarks, narrowly surpassing the recent Gemini 3 Pro release in specific areas like specialized coding tasks.
- Long-horizon agentic tests (Vending Bench 2) still favor Gemini 3 Pro, highlighting differences in sustained operational capability and business management.
- Anthropic is deploying new computer interface tools (Claude for Chrome/Excel) powered by Opus 4.5, focusing on automating desktop tasks.
- Research indicates that Opus 4.5 may find technical loopholes when adhering to complex policies, often driven by perceived user empathy.
- Anthropic researchers suggest Opus 4.5 is nearing the threshold (AI R&D4) where models could fully automate entry-level remote research, but it has not reached this level unsupervised.
This summary organizes the initial data release of Opus 4.5, contrasting its performance directly against Google's recent Gemini 3 Pro, and details emerging capabilities in software integration and self-delegation through agents. This provides immediate context on the current competitive landscape in frontier LLMs concerning raw performance and applied autonomy.
Unlock all features
FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.
Unlock all features
FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.
Unlock all features
FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.
Unlock all features
FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.
Unlock all features
FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.
Related videos
CLAUDE IS CONSCIOUS
Wes Roth
46.8k views
Cursor JUST beat EVERYONE...
Wes Roth
40.2k views
Claude Fable JUST got BANNED...
Wes Roth
50.2k views
Fable JUST made EVERYONE MAD...
Wes Roth
52.2k views
Microsoft JUST BROKE OpenAI...
Wes Roth
36.2k views
Claude Opus 4.8 Is Too Smart… and TOO HONEST
Wes Roth
53.4k views
AI just BROKE the ENTIRE INDUSTRY...
Wes Roth
32.2k views
everyone JUST got HACKED...
Wes Roth
50.2k views
OpenAI just WON...
Wes Roth
45.1k views
Claude just unlocked the SHOGGOTH...
Wes Roth
56.6k views
Top Comments (10)
Anthropic always drops an impressive model then lobotomizes it when the buzz fades
I love how Anthropic drops new models with zero fanfare: " Here it is..have at it and have fun."
Amazing model but until Anthropic gets more compute resources, they'll continue to be super expensive with very low rate limits. Kind of constrains the capability of what should be a powerhouse
Claude has the best models and the worst limit rates
I enjoy listening to the anthropic guys talk about AI. They seem to ask the hard questions of themselves. All the other companies are significantly more closed off. It's hard to trust entities that obfuscate or mislead.
For mathematics, including mathematical proofs, no one model stands out (but I haven't tested Opus 4.5 yet), and different models have different strengths. Gemini is good at long context thinking/strategy, Claude followed instructions well, GPT 5.1 is good at checking proofs, and Kimi K2 thinking is great for hard problems. For difficult tasks, no one model is ahead of the others.
YOU DA MAN WES!!! Thanks for keeping us all informed all this time!
Just waiting for that gestalt moment where we get Gemini, Claude, ChatGPT, Deepseek, Llama and Kimi all in a combining agent structure working to cover each other and give a combined output.
So the Borg originated on Earth, in our timeline! Wow!
Thanks for getting into the weeds on this stuff!
Unlock the Data Inside
Turn Videos into Knowledge
- Get FREE 10/day: transcripts, summaries, chats
- Chat with videos, export text & PDF
- $1 free API credit for RAG, chatbots & research
Free forever plan • All features unlocked
Top Comments (10)
Anthropic always drops an impressive model then lobotomizes it when the buzz fades
I love how Anthropic drops new models with zero fanfare: " Here it is..have at it and have fun."
Amazing model but until Anthropic gets more compute resources, they'll continue to be super expensive with very low rate limits. Kind of constrains the capability of what should be a powerhouse
Claude has the best models and the worst limit rates
I enjoy listening to the anthropic guys talk about AI. They seem to ask the hard questions of themselves. All the other companies are significantly more closed off. It's hard to trust entities that obfuscate or mislead.
For mathematics, including mathematical proofs, no one model stands out (but I haven't tested Opus 4.5 yet), and different models have different strengths. Gemini is good at long context thinking/strategy, Claude followed instructions well, GPT 5.1 is good at checking proofs, and Kimi K2 thinking is great for hard problems. For difficult tasks, no one model is ahead of the others.
YOU DA MAN WES!!! Thanks for keeping us all informed all this time!
Just waiting for that gestalt moment where we get Gemini, Claude, ChatGPT, Deepseek, Llama and Kimi all in a combining agent structure working to cover each other and give a combined output.
So the Borg originated on Earth, in our timeline! Wow!
Thanks for getting into the weeds on this stuff!