GPT‑5.4 mini & nano
What happened
OpenAI released GPT‑5.4 mini and nano — smaller models that deliver near‑flagship performance at much lower latency and cost, enabling high‑volume real‑time tasks. That tech shift makes inexpensive, always‑on automation for agent comms and lead triage materially more viable. (zdnet.com)
Why it matters
OpenAI published GPT‑5.4 mini and nano on March 17, 2026 as distinct, smaller variants of the GPT‑5.4 family. (openai.com)) On public benchmarks OpenAI listed, GPT‑5.4 scored 57.7% on SWE‑Bench Pro while GPT‑5.4 mini scored 54.4% and GPT‑5.4 nano scored 52.4%, showing a 3–5 point gap to the flagship on that metric. (openai.com)) OpenAI reported GPT‑5.4 mini runs more than 2× faster than the prior GPT‑5 mini and named classification, data extraction, ranking, and coding subagents as primary target workloads. (openai.com)) GPT‑5.4 mini is available via the API, Codex, and ChatGPT with a 400,000‑token context window and OpenAI‑published prices of about $0.75 per million input tokens and $4.50 per million output tokens. (thenewstack.io)) GPT‑5.4 nano is an API‑only option positioned as OpenAI’s cheapest high‑throughput model at roughly $0.20 per million input tokens and $1.25 per million output tokens, intended for ultra‑low‑latency classification and extraction tasks. (thenewstack.io)) OpenAI and partners describe a multi‑model “planner + subagent” pattern where GPT‑5.4 handles planning and mini/nano run parallel, focused subtasks, and OpenAI says mini consumes about 30% of a GPT‑5.4 quota in Codex for routine code work. (techcommunity.microsoft.com))
Key numbers
- OpenAI released GPT‑5.4 mini and nano — smaller models that deliver near‑flagship performance at much lower latency and cost, enabling high‑volume real‑time tasks.
- (zdnet.com) OpenAI published GPT‑5.4 mini and nano on March 17, 2026 as distinct, smaller variants of the GPT‑5.4 family.
- (openai.com)) On public benchmarks OpenAI listed, GPT‑5.4 scored 57.7% on SWE‑Bench Pro while GPT‑5.4 mini scored 54.4% and GPT‑5.4 nano scored 52.4%, showing a 3–5 point gap to the flagship on that metric.
- (openai.com)) OpenAI reported GPT‑5.4 mini runs more than 2× faster than the prior GPT‑5 mini and named classification, data extraction, ranking, and coding subagents as primary target workloads.
What happens next
- (openai.com)) OpenAI reported GPT‑5.4 mini runs more than 2× faster than the prior GPT‑5 mini and named classification, data extraction, ranking, and coding subagents as primary target workloads.
Quick answers
What happened in GPT‑5.4 mini & nano?
OpenAI released GPT‑5.4 mini and nano — smaller models that deliver near‑flagship performance at much lower latency and cost, enabling high‑volume real‑time tasks. That tech shift makes inexpensive, always‑on automation for agent comms and lead triage materially more viable. (zdnet.com)
Why does GPT‑5.4 mini & nano matter?
OpenAI published GPT‑5.4 mini and nano on March 17, 2026 as distinct, smaller variants of the GPT‑5.4 family. (openai.com)) On public benchmarks OpenAI listed, GPT‑5.4 scored 57.7% on SWE‑Bench Pro while GPT‑5.4 mini scored 54.4% and GPT‑5.4 nano scored 52.4%, showing a 3–5 point gap to the flagship on that metric. (openai.com)) OpenAI reported GPT‑5.4 mini runs more than 2× faster than the prior GPT‑5 mini and named classification, data extraction, ranking, and coding subagents as primary target workloads. (openai.com)) GPT‑5.4 mini is available via the API, Codex, and ChatGPT with a 400,000‑token context window and OpenAI‑published prices of about $0.75 per million input tokens and $4.50 per million output tokens. (thenewstack.io)) GPT‑5.4 nano is an API‑only option positioned as OpenAI’s cheapest high‑throughput model at roughly $0.20 per million input tokens and $1.25 per million output tokens, intended for ultra‑low‑latency classification and extraction tasks. (thenewstack.io)) OpenAI and partners describe a multi‑model “planner + subagent” pattern where GPT‑5.4 handles planning and mini/nano run parallel, focused subtasks, and OpenAI says mini consumes about 30% of a GPT‑5.4 quota in Codex for routine code work. (techcommunity.microsoft.com))
