OpenAI fixes ‘lazy’ GPT-4 Turbo model and drops pricing

OpenAI slashes prices for GPT-3.5 Turbo and boosts GPT-4 Turbo performance.

OpenAI reveals new improvements to its AI models and updated pricing. (Image: Reuters)

OpenAI has released several updates to its AI models and pricing, marking the latest in a series of incremental improvements by the research company. The changes include lower pricing for its popular GPT-3.5 Turbo model, a performance boost for GPT-4 Turbo, and upgraded text embedding models.

Most notably, OpenAI has cut the per-token pricing of GPT-3.5 Turbo by 50% for input and 25% for output. This model powers applications like ChatGPT and serves as an industry benchmark for conversational AI. Lower prices make the API more accessible for developers building text-intensive products that require analysing documents or books. With open-source models catching up to OpenAI, reduced pricing helps retain customers.

GPT-3.5 Turbo also gains a new version 0125 with unspecified “improvements.” Since the last iteration was 0613, some may have expected more details on advancements from OpenAI.

For GPT-4 Turbo, a new preview model 0125 arrives that OpenAI says “completes tasks like code generation more thoroughly” to reduce “laziness” where the model avoids finishing work. This fix aims to boost performance on technical tasks compared to previous versions. GPT-4 with vision capabilities remains in preview but will launch more broadly in coming months.

On the technical side, upgraded text embedding models provide better semantic representations of language for researchers and engineers.

While not earth-shattering, OpenAI’s latest tweaks reflect the company’s iterative approach to enhancing its AI lineup.