gpt-4.1-mini-2025-04-14

GPT-4.1 mini model, version 2025-04-14

GPT‑4.1 mini is the “compact powerhouse” of the GPT‑4.1 family. It preserves much of GPT‑4.1’s intelligence while dramatically reducing cost and latency, making it better suited as a scaled, default workhorse. Both GPT‑4.1 and GPT‑4.1 mini support up to 1M tokens of context and share a June 2024 knowledge cutoff; however, GPT‑4.1 targets peak performance, whereas GPT‑4.1 mini is optimized for price‑performance and responsiveness.

On academic and general benchmarks such as MMLU and GPQA, GPT‑4.1 mini scores only slightly below GPT‑4.1 while already surpassing GPT‑4o and GPT‑4o mini overall. On instruction‑following benchmarks (MultiChallenge, IFEval, internal hard instruction‑following sets), it trails GPT‑4.1 by a small margin, yet is strong enough for most production use cases that require strict formats, complex constraints, and multi‑turn dialog consistency. Compared with GPT‑4.1, its absolute ceiling on top‑end coding tasks like SWE‑bench is lower, but it still performs very well for everyday coding assistance, SQL help, and lightweight code review.

The real differentiator is efficiency. In the API, GPT‑4.1 mini’s pricing is roughly one‑fifth to one‑sixth of GPT‑4.1 ($0.40 vs $2.00 input, $1.60 vs $8.00 output per million tokens) and, in practice, it almost halves latency. This makes it ideal for high‑traffic, cost‑sensitive applications that still need strong intelligence, such as in‑product assistants, bulk document processing, and analytics copilots. Overall, GPT‑4.1 mini strikes a compelling balance between “near‑flagship quality” and aggressive cost efficiency.