OpenAI Launches GPT-5.4 nano and mini — Cheapest, Most Capable Small Models Yet

OpenAI has launched two new, cost-efficient versions of its flagship GPT-5.4: GPT-5.4 nano and GPT-5.4 mini. The company calls them "our most capable small models ever," with benchmarks showing they approach full GPT-5.4 performance on coding and agentic tasks — at a fraction of the price.

What's New?

GPT-5.4 (full version) launched on March 5, 2026 with computer use capabilities, a 1 million token context window, and strong reasoning. OpenAI now completes the family with two specialized variants:

• GPT-5.4 nano — ultra-fast, extremely affordable, designed for high-volume tasks
• GPT-5.4 mini — balances capability and cost, near full model performance on coding

Both support Codex integration and the Tool Search engine that lets models automatically identify and use the right tools without developers listing them in every prompt.

Impressive Benchmarks

GPT-5.4 (full) delivers:

• 75% score on OSWorld — beats humans at desktop navigation (72.4%)
• 33% fewer hallucinations than GPT-5.2
• 47% fewer tokens used on many tasks compared to older models
• Matches professionals across 44 occupations (83% level)

Nano and mini aim to match these numbers on coding and agentic tasks.

What This Means for Developers

Lower costs at scale. For enterprises running AI agents against large datasets — such as e-commerce, customer service, or document processing — nano/mini can dramatically cut API expenses while maintaining high performance.

OpenAI positions this as the answer to competition from Anthropic Claude Haiku and Google Gemini Flash, which have dominated the affordable segment.

Availability

Available via ChatGPT and OpenAI API starting today. Enterprise customers with Codex access get full integration immediately.