Will an LLM with less than 10B parameters beat GPT4 by EOY 2025?
Mini
15
Ṁ1812
2026
98%
chance

How much juice is left in 10B parameters?

the original GPT4-0314 (ELO 1188)

judged by Lmsys Arena Leaderboard

current SoTA: Llama 3 8B instruct (1147)

Get
Ṁ1,000
and
S1.00
Sort by:
bought Ṁ500 YES

@mods Resolves as YES.. The creator's account is deleted, but Gemini 1.5 Flash-8B easily clears the bar (ELO higher than 1188) with an ELO of 1212, see https://lmarena.ai/

bought Ṁ400 YES

The market creator is banned, but it seems the YES resolution criterion has been met because Gemma-2-9B-it is at elo 1188, above GPT-4-0314 at 1886. However, one could argue it needs to be 1189 or higher because GPT-4-0314 was at 1188 when the market was created.

bought Ṁ10 YES

Open source models have an inherent advantage in that they don't need to conform to censorship/"safety" policies, making them much more useful and thus getting them a higher ELO. Having to conform to policies is a major handicap for the larger closed-source models.

@singer There is a without refusal board on lmsys. The disparity between main and that is how much advantages not having censorship’s give u