gpt-5.4-nano-2026-03-17

Nano-tier latency suits high-volume first-line support replies on simple queries. Escalation and multi-turn reasoning quality are not documented, and OpenAI's US-jurisdiction exposure applies to customer data routing.

Short-form summarisation of transcripts is within the nano-tier's expected range. Long meeting transcripts may exceed a context window that is not published.

Adequate for short-to-medium summarisation where speed and cost matter more than depth. Context window is unspecified, so long-input summarisation cannot be confirmed.

Nano-tier model can handle schema-bound extraction from short, well-formed inputs. Complex or long-document extraction is not supported by any published context window or benchmark.

Nano-tier size class is suited to lightweight, high-volume tagging workloads where label accuracy needs are modest. No published accuracy figures are available to confirm performance against larger variants.

Low-latency nano-tier profile fits real-time intent routing ahead of heavier models. Routing accuracy is unverified in the card, so this is a provisional fit.

Nano-tier size class generally trades multi-step reasoning depth for speed, which works against multi-file agentic coding loops. No coding benchmark or tool-use capability is documented to offset this.

Closed-weights model under US CLOUD Act jurisdiction, which is a poor match for compliance-driven redaction of sensitive data. Redaction recall is also not documented for a nano-tier variant.