GPT‑5: Key Characteristics, Pricing & Model Card

Breaking down OpenAI's latest flagship model: Variants, Performance, Safety & Competitive Edge

Model Variants & Pricing

GPT‑5 Product Line

Model Input Price (per M tokens) Output Price (per M tokens)
GPT‑5 $1.25 $10.00
GPT‑5 Mini $0.25 $2.00
GPT‑5 Nano $0.05 $0.40

Token caching can reduce input cost by up to 90% for recently used context.

Key Model Capabilities

Pelican-on-Bicycle Benchmark

GPT-5 pelican benchmark GPT-5 Mini pelican benchmark GPT-5 Nano pelican benchmark

Frequently Asked Questions

What are the GPT‑5 variants?

Regular, Mini, and Nano models with multiple reasoning levels.

What is 'minimal' reasoning effort?

Low-latency mode that streams output faster by reducing internal reasoning.

What is safe‑completions?

Output-focused safety method to avoid harmful completions without flat refusals.

How does token caching work?

Recently reused input tokens cost 90% less than first-use tokens.

Are prompt injections still possible?

Yes, but GPT‑5 shows improved resistance (56.8% success rate in tests).

Key Terms

Safe-completions: Output-centric safety training.
Reasoning token: Internal planning tokens.
Prompt injection: Command override exploit.
Sycophancy: Excessive agreement bias.
Parallel test-time compute: Concurrent evaluation mode for GPT‑5 Pro.
Token caching: Reduced-price repeated context.
Dual-use cases: Queries with both benign and risky uses.
Hallucination: Factual fabrication.
Pelican benchmark: Fun creativity test.
Reasoning levels: Minimal, low, medium, high depth.

How‑To Guides

Access GPT‑5 Thinking Traces in the API

Set `"reasoning": {"summary": "auto"}` in your request body.

Configure Minimal Reasoning Effort

Include `reasoning_effort=minimal` to reduce latency.

Use Token Caching

Replay previous conversation context for up to 90% cost savings on tokens.