Skip to content
Trending storyDeveloping

Allie K. Miller compares GPT-4 with Astra on context window and GPQA scores

1 article1 sourcesince Oct 17Last article 1h ago ·

Overview

AISummary of 1 article

Allie K. Miller compares GPT-4 with a model she calls Astra, in a post on X. She says GPT-4 had an 8k or 32K token context cap and scored 39% on GPQA Diamond, and that its users were limited to 50 messages every 4 hours.

She says Astra has a 1M token context window and scores 96% on GPQA Diamond.

The post gives no source or test setup for the Astra figures, and the report does not confirm them independently.

Written by AI from the articles below · updated Oct 17, 1:03 PM ET

Check the sources:

Latest developmentOct 17, 1:00 PM ET
AI model progress since GPT-4: context window and GPQA scores

Article timeline

The articles in this story. Times are ET.

Oct 17
  1. Allie K. MillerX
    AI model progress since GPT-4: context window and GPQA scores

    AIAllie K. Miller compares GPT-4 with Astra, noting GPT-4 had an 8k/32K token cap and scored 39% on GPQA Diamond. She says Astra has a 1M token context window and scores 96% on GPQA Diamond. She also recalls that GPT-4 users were limited to 50 messages every 4 hours.

    Image from @alliekmiller's post

Heat trend

Not enough continuous observations to show a trend yet.