OpenAI Unveils High-Performance Mini Models with Coding Boosts on Free Tier
On March 17–18, OpenAI launched its most capable small models yet—the GPT‐5.4 mini and nano variants—designed to deliver near-flagship performance at significantly lower costs for high-volume workloads like coding agents and automated workflows; the new free tier of ChatGPT now grants access specifically to these improved versions which feature enhanced reasoning, multimodal understanding capabilities derived from their larger counterparts while trading some accuracy for speed.
Key Points
-
1OpenAI has launched two new compact AI models the 'mini' version (available to free users) and a smaller/nano variant.
-
2The primary focus of these releases is speed, efficiency, and cost reduction for high-volume workloads like coding assistants and automation agents.
-
3While trading some accuracy found in larger flagship GPT-5.4 models, the mini model offers near-flagship performance at significantly lower latency (over twice as fast) to save costs on API usage.
Developments
OpenAI launched compact models named "GPT-5.4 Mini" and "Nano," which are available to Free users on ChatGPT with speeds up twice as fast as GPT-5 mini while maintaining high benchmark scores like 88% accuracy in the GPQA Diamond test. These small-scale AI subagents can handle specific tasks such as coding or document review, allowing larger flagship models to focus solely on complex planning and final judgment within a hierarchical workflow used by platforms like Notion.
OpenAI announced new "mini" and "nano" versions of the GPT-5.4 model designed for faster speeds at lower costs while maintaining high performance in coding tasks like debugging front-end generation, with pricing set between $0.20 per million input tokens to be accessed via API or ChatGPT (except nano).
OpenAI has launched two new versions of its flagship model, "GPT 5" (referred to as the family's base), which are designed for high-volume and latency-sensitive applications. The GTP-4 mini offers improved performance across coding and reasoning while running over twice faster than previous models at lower cost; it is available via API on Codex or ChatGPt, whereas "GPT 5" nano provides maximum efficiency specifically through the API