OpenAI released GPT-5.4 mini and GPT-5.4 nano on March 17, bringing its small-model lineup up to speed with the flagship GPT-5.4 that shipped two weeks prior. The pitch: near-flagship performance at dramatically lower cost and latency. The announcement post frames them as purpose-built for coding assistants, subagents, and anything where waiting half a minute for a response kills the product experience.
Mini is the headliner. It runs over 2x faster than GPT-5 mini and scores 54.4% on SWE-Bench Pro, close to the full GPT-5.4's 57.7%, per OpenAI's own benchmarks. On OSWorld-Verified, which measures computer-use tasks, mini hit 72.1% against the flagship's 75.0%. Those are company-reported numbers, so take them accordingly. Perplexity's deputy CTO Jerry Ma offered a more tempered read, calling mini's reasoning "strong" while noting nano works best for "live conversational workflows."
Nano is the real cost play: $0.20 per million input tokens, $1.25 per million output. That undercuts even Google's Gemini 3.1 Flash-Lite, as Simon Willison noted. OpenAI pitches it for classification, extraction, ranking, and lightweight coding subagents. It's API-only, no ChatGPT access.
Mini is live in ChatGPT, Codex, the API, and already rolling out in GitHub Copilot. Free and Go ChatGPT users get it through the Thinking feature. Paid users get it as a rate-limit fallback for GPT-5.4 Thinking. In Codex, mini burns only 30% of the GPT-5.4 quota.
Bottom Line
GPT-5.4 mini scores within 3 percentage points of the flagship on SWE-Bench Pro while running at more than double the speed, and nano undercuts most competitors on price at $0.20 per million input tokens.
Quick Facts
- GPT-5.4 mini: 2x faster than GPT-5 mini (company-reported)
- GPT-5.4 mini SWE-Bench Pro: 54.4% vs flagship's 57.7% (company-reported)
- GPT-5.4 nano pricing: $0.20/1M input tokens, $1.25/1M output tokens
- Mini available in ChatGPT, Codex, API, GitHub Copilot
- Nano is API-only
