OpenAI has rolled out an upgraded version of its GPT-4 Turbo large language model to ChatGPT’s paid versions.

The company announced the update in a late Thursday post on X. “Our new GPT-4 Turbo is now available to paid ChatGPT users,” OpenAI detailed. “We’ve improved capabilities in writing, math, logical reasoning, and coding.”

GPT-4, a successor to the model that powered ChatGPT when the latter service launched, debuted last March. Compared with its predecessor, the fourth-generation LLM is better at reasoning tasks and less likely to output erroneous information. Six months ago, OpenAI introduced an upgraded version called GPT-4 Turbo that delivered another increase in response quality.

The company has since released multiple improvements to the model. The version that rolled out to ChatGPT’s paid tiers this week, gpt-4-turbo-2024-04-09, is the most recent iteration.

The new model was trained on more recent data than its predecessor. According to OpenAI, the previous iteration of GPT-4 Turbo only had access to information generated before April 2023. In contrast, the latest version has access to data created as recently as December. 

OpenAI compared the two models using Simple Evals, an internally developed benchmarking tool it has open-sourced on GitHub. The tool evaluates the accuracy of an LLM’s output by having it tackle six collections of questions spanning dozens of topics. According to OpenAI, gpt-4-turbo-2024-04-09 performed better than its predecessor across all six question sets.

OpenAI observed the biggest improvement when it tested the new LLM on an evaluation dataset called GPQA. Developed by Google LLC, the file contains 448 graduate-level multiple-choice questions spanning topics such as math and physics. The latest iteration of GPT-4 Turbo answered 49.1% of the questions accurately, a nearly 10% improvement over the model’s previous version.

The new LLM delivered a similar accuracy increase in a separate benchmark test that involves math questions. Across the four other question collections OpenAI evaluated, the model demonstrated improvements of a few percentage points.

OpenAI says that gpt-4-turbo-2024-04-09 is accessible to users of ChatGPT’s paid Plus, Team and Enterprise tiers. The former edition is designed for consumers while the latter two are geared towards businesses. ChatGPT Enterprise, the most recent addition to the lineup, offers cybersecurity controls not available in the other versions along with faster response times.

Reuters reported today that OpenAI Chief Executive Officer Sam Altman recently hosted a series of events to promote ChatGPT Enterprise. According to the news agency’s sources, the events were attended by more than 300 executives from Fortune 500 companies. It’s believed the consumer version of ChatGPT is already used by employees at more than 90% of the Fortune 500. 

Image: Unsplash