OpenAI's latest model promises reasoning leaps, lower hallucination rates, and multimodal improvements. But how does it hold up against the hype? We break down the real changes, benchmark numbers, and what it means for developers building on the API.
GPT-5 Is Officially Live
OpenAI released GPT-5 to ChatGPT Plus and API users, billing it as its most capable model to date. The rollout is staged — free-tier users get access within weeks.
- Available now for ChatGPT Plus and Team plans
- API access via gpt-5 model string
- Free tier rollout expected within 4–6 weeks
What's Actually Different
OpenAI claims a significant jump in reasoning benchmarks and a notable reduction in confabulation — the model's tendency to state falsehoods confidently.
- MMLU score: 92.4% vs GPT-4o's 88.7%
- Hallucination rate down by an estimated 38% on TruthfulQA
- Native image, audio, and document input in a single call
"GPT-5 is not just an incremental update — it represents a fundamental shift in how the model approaches multi-step reasoning tasks." — OpenAI blog
API Pricing Changes
Despite increased capability, OpenAI has dropped pricing compared to GPT-4o at launch — a deliberate move to accelerate adoption.
- Input: $2.50 / 1M tokens
- Output: $10.00 / 1M tokens
- Batch API discount: 50% off
Should You Migrate?
For most production workloads the answer is yes, but test your evals first. Some edge cases around structured output formatting have regressed from GPT-4o.
- Run your existing eval suite before switching
- Watch for JSON mode regressions reported by early adopters
- Context window: 256K tokens — up from 128K
