OpenAI has officially unveiled GPT‑5, the latest large‑language model powering ChatGPT. The release marks the first major upgrade since GPT‑4’s debut in March 2023. CEO Sam Altman says GPT‑5 is like “having a team of PhD‑level experts in your pocket.”.
Why GPT‑5 Matters
Smarter, Faster, Safer
GPT‑5 is more accurate than its predecessors and markedly less prone to hallucinations. Early benchmarks show outstanding gains in coding, writing and health‑related tasks.
Available to Everyone
The model is rolling out to all 700 million weekly ChatGPT users. A new internal router automatically selects between GPT‑5 variants (standard, mini, nano, pro, thinking) based on query complexity.
Massive Context Window
GPT‑5 handles 256 000 tokens, enabling much longer conversations or document ingestion without losing context.
Feature Highlights
| Feature | Description |
|---|---|
| Test‑time compute router | Switches to a deeper‑thinking variant when a query demands it. |
| Vibecoding / software on demand | Generates complete apps or websites from natural‑language prompts. |
| Personality themes | Users can choose from four preset personas: Cynic, Robot, Listener, Nerd. |
| Mini & Nano models | Lower‑cost, lower‑latency versions for everyday tasks. |
| New API pricing | $1.25 / $10 per M input/output tokens (GPT‑5); cheaper for mini/nano. |
| Productivity connectors | Gmail, Google Calendar and Contacts integrations (Pro tier). |
| Health expertise | Top scores on medical reasoning benchmarks. |
| Safe completions | Fewer refusals and clearer explanations when boundaries are hit. |
What GPT‑5 Can Do
Coding & Software Development
Hailed as the best OpenAI model for programming, GPT‑5 tops coding benchmarks like SWE‑Bench Verified and Aider Polyglot, generating or refactoring complex codebases on demand.
Writing & Productivity
The model drafts essays, emails and long‑form documents while maintaining coherence across thousands of tokens.
Health‑Related Queries
GPT‑5 provides more reliable answers to medical questions and proactively flags potential concerns.
Natural Conversation
Users describe GPT‑5 as more fluid and human‑like than any prior version.
Accessibility & Pricing Tiers
- Free: GPT‑5 + GPT‑5‑mini with usage limits.
- Plus (€20/mo): Higher limits on GPT‑5/mini.
- Pro (€200/mo): Unlimited GPT‑5 and exclusive access to GPT‑5‑pro & GPT‑5‑thinking.
- API: Tiered pricing for gpt‑5, gpt‑5‑mini, gpt‑5‑nano.
Limitations & Criticisms
While GPT‑5 represents a leap in capability, it still can’t learn after deployment, and OpenAI hasn’t disclosed its training data set, raising transparency concerns.
Performance Comparison (o1 · o3 · GPT‑4o · GPT‑4.1 · GPT‑5)
Below is a side‑by‑side look at key benchmarks across generations.
| Model | SWE‑Bench Verified (%) | HealthBench Hard (%) | Hallucination Rate (%) |
|---|---|---|---|
| GPT‑o1 | 48.9 | ~25.0 | ~38.0 |
| GPT‑o3 | 71.7 | 31.6 | ~24.0 |
| GPT‑4o | 33.2 | 31.6 | 4.5 |
| GPT‑4.1 | 54.6 | ~40.0 | 2.2 |
| GPT‑5 | 74.9 | 46.2 | 0.7 |
Key Takeaways
- Coding: GPT‑5 edges past o3 and towers over GPT‑4o in real‑world coding tasks.
- Health: GPT‑5 delivers the highest score yet on HealthBench Hard.
- Safety: Hallucinations drop sharply—from ~24 % in o3 to < 1 % in GPT‑5.
Implications for the Future
GPT‑5’s arrival signals a maturing AI ecosystem poised for deeper enterprise adoption, especially in software development, healthcare and knowledge work. Responsible deployment and transparency remain essential as these models grow ever more capable.