Will Llama 3 405B match or surpass release-time GPT 4 on vibebench?
11
Ṁ1kṀ2.1kresolved Aug 31
Resolved
YES1H
6H
1D
1W
1M
ALL
The public basically never got a full picture of GPT4 capabilities. As a result of quantization/distillation/expert routing tricks speeding it up, plus whatever safety ablations got applied, it's common consensus that March 2023 GPT4 and whatever Microsoft exposed via Sydney-era Bing was significantly better than what's accessible today.
This market resolves based on people's Twitter/Discord vibes comparing 405B (instruct or base) against what they remember from then.
This question is managed and resolved by Manifold.
Market context
Get 
1,000 to start trading!
1,000🏅 Top traders
| # | Trader | Total profit |
|---|---|---|
| 1 | Ṁ423 | |
| 2 | Ṁ162 | |
| 3 | Ṁ88 | |
| 4 | Ṁ11 | |
| 5 | Ṁ11 |
Sort by:
https://x.com/repligate/status/1824179931057930617
Hermes-405B finetune in particular seems to have hit the same category of weird as Sydney did.
People are also trading
Related questions
Llama 5 outperforms GPT 4o on LM Arena?
85% chance
When will an open-source LLM be released with a better performance than GPT-4?
GPT-4 performance and compute efficiency from a simple architecture before 2026
19% chance
In what year will a GPT4-equivalent model be able to run on consumer hardware?
2026
In what year will a GPT4-equivalent model be able to run on consumer hardware?
2026