Content
summary Summary

OpenAI plans to release its new AI model, codenamed "Strawberry," within the next two weeks. The model, designed to enhance logical reasoning capabilities, will be integrated into ChatGPT.

Ad

This information comes from two individuals who have tested the model, according to The Information. The launch appears to be ahead of the previously reported fall release.

Strawberry's "thinking time" sets it apart

Strawberry's key feature is a 10 to 20-second "thinking" period before responding to queries. This pause aims to help the model tackle complex problems in areas like design, mathematics, and programming, even for tasks it hasn't encountered before.

The model reportedly uses a specialized form of "post-training" to adapt pre-trained models for specific tasks. This approach could eliminate the need for additional prompts or workarounds that users currently employ to get more advanced results from ChatGPT.

Ad
Ad

In an internal test, Strawberry is said to have scored over 90% on a collection of mathematical mastery tasks. Strawberry is also said to be providing training examples for OpenAI's GPT-4 successor, codenamed "Orion".

A berry good addition to ChatGPT?

While Strawberry will be part of ChatGPT, it's designed as a separate offering with some differences. The initial version will only process text, unlike other current OpenAI models that can handle multiple types of input like images or files.

The company intends to use Strawberry to support software and AI developers, enable autonomous internet searches with predictive planning, and facilitate in-depth research, among other applications.

However, early testers have noted some issues. The model sometimes takes too long to process even simple questions and isn't yet fully adapted to certain user preferences, such as preferred code formats. Some initial testers felt that Strawberry's slightly improved answers compared to GPT-4o didn't justify the additional 10 to 20-second wait time, The Information reports.

The Information speculates that pricing for Strawberry could differ from ChatGPT and may include rate limits. A higher-priced subscription might be available for faster responses. OpenAI is reportedly considering significantly higher prices for its more advanced AI models overall.

Ad
Ad
Join our community
Join the DECODER community on Discord, Reddit or Twitter - we can't wait to meet you.
Recommendation
Support our independent, free-access reporting. Any contribution helps and secures our future. Support now:
Bank transfer
Summary
  • OpenAI is set to release "Strawberry," a new AI model focusing on logical reasoning, as part of ChatGPT within the next two weeks. The details of its integration and pricing structure are still not fully clear.
  • Strawberry's main feature is a 10-20 second "thinking" period before it responds to queries. The model uses specialized post-training techniques to tackle complex math and programming problems, aiming to improve upon current ChatGPT capabilities.
  • Some testers found that the slight improvements over GPT-4o didn't justify the extended response time.
Sources
Online journalist Matthias is the co-founder and publisher of THE DECODER. He believes that artificial intelligence will fundamentally change the relationship between humans and computers.
Join our community
Join the DECODER community on Discord, Reddit or Twitter - we can't wait to meet you.