OpenAI slashes fine-tuning costs for GPT-4o mini with special limited-time offer

Compared to GPT-3.5 Turbo, GPT-4o mini is also 90 percent cheaper for input tokens and 80 percent cheaper for output tokens. The training context is four times longer than GPT-3.5 Turbo at 65,000 tokens, while the inference context is eight times longer at 128,000 tokens.

The new GPT-4o mini model also offers more performance and supports image processing, although fine-tuning is currently limited to text. Fine-tuning is available to OpenAI level 4 and 5 developers (minimum $250 paid and 14+ days since first successful payment) through the Fine-Tuning dashboard.

At the same time, Microsoft has introduced serverless fine-tuning services for the Phi 3-mini and Phi 3-medium models in Azure AI. According to Microsoft, this allows developers to quickly tune models for cloud and edge scenarios without requiring extensive computing resources.

Microsoft also reports "significant improvements in core quality, instruction-following, and structured output" through updates for Phi-3-mini.

LLM market sees increasing price pressure

OpenAI's discount campaign also highlights the changing language model market: More models from more providers, some open source, are competing for a limited number of application scenarios in a market that generates significant interest but little revenue.

Competition is currently intensifying for both small and large models, which have seen only incremental improvements since the introduction of the GPT-4 class. The result is a price war.