Hacker News new | ask | show | jobs
The Claude 3 Model Family: Opus, Sonnet, Haiku [pdf] (www-cdn.anthropic.com)
33 points by 1wheel 833 days ago
2 comments

Impressive benchmarks here. The 90% eval for one of the math categories on 0-shot vs 74.5% GPT-4 8-shot is nice.

https://twitter.com/AnthropicAI/status/1764653830468428150

Nice. GPT-4 beater finally perhaps (until 5 is launched which I guess must be pretty soon now)
Related ongoing thread:

Claude 3 model family - https://news.ycombinator.com/item?id=39590666 - March 2024 (347 comments)