Curiosity · AI Model

GPT-5 mini

GPT-5 mini is the mid-tier of OpenAI's GPT-5 family, released alongside the flagship in August 2025. It inherits GPT-5's unified router (automatic think-vs-chat switching) and tool-use stability, but runs at a fraction of the cost — making it the default pick for production chat, agents, and RAG pipelines where full GPT-5 is overkill.

Model specs

Vendor
OpenAI
Family
GPT-5
Released
2025-08
Context window
400,000 tokens
Modalities
text, vision, code
Input price
$0.25/M tok
Output price
$2/M tok
Pricing as of
2026-04-20

Strengths

  • Unified model handles easy chat and harder reasoning without model switching
  • Excellent cost-per-quality — roughly 5x cheaper than flagship GPT-5
  • Strong tool-use reliability inherited from GPT-5 training
  • Large 400K context suffices for most real-world documents

Limitations

  • Trails flagship GPT-5 on frontier reasoning, research-grade code
  • Vision quality below top-tier for dense document OCR
  • Effort-routing can occasionally over-think simple prompts, adding latency

Use cases

  • High-volume chatbots and support agents
  • RAG pipelines where cost per query is the constraint
  • Content generation and summarisation at scale
  • Tool-calling agents where full GPT-5 is cost-prohibitive

Benchmarks

BenchmarkScoreAs of
SWE-bench Verified≈65%2025-10
AIME 2024 (math)≈85%2025-10
MMLU-Pro≈80%2025-10

Frequently asked questions

What is GPT-5 mini?

GPT-5 mini is the mid-tier of OpenAI's GPT-5 model family, released in August 2025. It uses the same unified reasoning-and-chat architecture as flagship GPT-5 but at roughly one-fifth the price, making it the default choice for production deployments.

When should I use GPT-5 mini versus GPT-5?

Use GPT-5 mini for high-volume production workloads — chat, RAG, standard agents — where cost matters. Use flagship GPT-5 for frontier reasoning, complex coding agents, and cases where quality outranks cost.

How much does GPT-5 mini cost?

As of April 2026, GPT-5 mini is priced around USD 0.25 per million input tokens and USD 2 per million output tokens — roughly five times cheaper than flagship GPT-5.

Does GPT-5 mini support reasoning?

Yes. GPT-5 mini uses the same unified reasoning architecture as flagship GPT-5, automatically deciding when to engage extended thinking based on prompt difficulty. You can also force reasoning effort via API parameters.

Sources

  1. OpenAI — Introducing GPT-5 — accessed 2026-04-20
  2. OpenAI — Pricing — accessed 2026-04-20