OpenAI released GPT-5.4 mini and GPT-5.4 nano today — two small models designed for latency-sensitive, cost-conscious workloads where the biggest model is rarely the right choice.

What's New

GPT-5.4 mini is OpenAI's headline small model release. It runs more than 2x faster than the previous GPT-5 mini while scoring significantly higher on coding and reasoning benchmarks. On SWE-Bench Pro, it achieves 54.4% compared to 45.7% for GPT-5 mini, and approaches GPT-5.4 performance on OSWorld-Verified (72.1% vs. 75.0%).

GPT-5.4 nano is the smallest and cheapest option — API-only, aimed at classification, data extraction, ranking, and simpler coding subagents.

Designed for Agentic Systems

Both models are explicitly built for agentic architectures. OpenAI highlights a pattern where a large model like GPT-5.4 handles planning and coordination while GPT-5.4 mini subagents execute narrower tasks in parallel — searching a codebase, reviewing files, or processing screenshots. The models also power computer use tasks, with mini approaching GPT-5.4 on multimodal benchmarks.

Pricing and Availability

GPT-5.4 mini is available today in the API, Codex, and ChatGPT. It has a 400k context window and costs $0.75 per 1M input tokens and $4.50 per 1M output tokens. In ChatGPT, it serves as the rate-limit fallback for paid users and the primary Thinking model for free and Go tiers.

GPT-5.4 nano is API-only for now. Pricing was not disclosed in the announcement.