Will Claude 3.5 Haiku be better than Claude 3 Opus? | Manifold

Will Claude 3.5 Haiku be better than Claude 3 Opus?

Basic

31

Ṁ4735

resolved Nov 5

Resolved

NO

1D

1W

1M

ALL

I will resolve it according to the most recent benchmark results from livebench.ai as soon as the results are available.

In the current benchmark Claude 3 Opus has a score of 50,03 on the 'Global average', while Claude 3.5 Sonnet scores 59,80.

This question is managed and resolved by Manifold.

#️ Technology

Get

1,000

and

3.00

Sort by:

sold Ṁ55 YES

bought Ṁ300 YES

https://x.com/AnthropicAI/status/1853498270724542658

3.5 Haiku looks comparable to 4o-mini (which is 7 points behind 3 Opus on LiveBench) and somewhat below 1.5 Flash (which is 0.5 points behind 3 Opus on LiveBench).

bought Ṁ30 NO

Ya, I'd consider it well below gemini-1.5-flash-002.

The GPQA scores are barely above Claude 3 Sonnet.

bought Ṁ10 YES

@JoshYou now it's overcorrected imo, I'm at 30-40%

Related questions

Will Claude 3.5 Opus be available via API by end of 2025?

Will Claude 3.5 Opus be able to draw me in tic-tac-toe while playing as O at least 1/3 of the time?

Does Claude 3.5 have control vector(s) to increase its capabilities?

Will Claude 3.5 Opus have a higher Chat Arena Elo than GPT-5?

How many parameters does the new possibly-SOTA large language model, Claude 3 Opus, have?

Will Claude 3.5 Opus beat OpenAI's best released model on the arena.lmsys.org leaderboard?

Is Claude 3.5 Sonnet a distilled or quantized version of a larger model?

What will Claude 3.5 Opus's reported 0-shot performance on GPQA Diamond be upon release?

Why is Claude 3.5 Sonnet such a good model for its size?

What will be the *first* ELO Rating of Claude 3.5 Opus in the LMSYS Arena?

Related questions

Will Claude 3.5 Opus be available via API by end of 2025?

Will Claude 3.5 Opus beat OpenAI's best released model on the arena.lmsys.org leaderboard?

Will Claude 3.5 Opus be able to draw me in tic-tac-toe while playing as O at least 1/3 of the time?

Is Claude 3.5 Sonnet a distilled or quantized version of a larger model?

Does Claude 3.5 have control vector(s) to increase its capabilities?

What will Claude 3.5 Opus's reported 0-shot performance on GPQA Diamond be upon release?

Will Claude 3.5 Opus have a higher Chat Arena Elo than GPT-5?

Why is Claude 3.5 Sonnet such a good model for its size?

How many parameters does the new possibly-SOTA large language model, Claude 3 Opus, have?

What will be the *first* ELO Rating of Claude 3.5 Opus in the LMSYS Arena?

© Manifold Markets, Inc.•Terms + Mana-only Terms•Privacy•Rules