OpenAI has officially unveiled GPT-5, its most powerful language model to date, marking what the company calls “a generational leap” in artificial intelligence reasoning, coding, and multimodal understanding. The new model outperforms its predecessor on virtually every benchmark, with particular gains in mathematics, scientific reasoning, and long-context comprehension.
What’s New in GPT-5
GPT-5 introduces a revamped reasoning architecture that allows the model to think through complex problems step by step before delivering a final answer. This so-called “extended thinking” mode, similar to features pioneered by competitors, significantly reduces hallucinations and improves accuracy on multi-step tasks.
OpenAI claims GPT-5 scores above 90th percentile on the bar exam, matches expert-level performance on medical licensing tests, and can write and debug production-ready code across dozens of programming languages. The model also demonstrates dramatically improved memory management for context windows exceeding 200,000 tokens.
Multimodal Capabilities
The new model natively handles images, audio, video, and documents in a single unified architecture. Users can upload a PDF, ask questions about a YouTube video, or have real-time voice conversations with near-human latency — all within the same session.
“We built GPT-5 to be the best assistant in the world, capable of handling any task you throw at it,” said OpenAI CEO Sam Altman in a statement accompanying the launch. “This is the model we’ve been working toward since the beginning.”
Availability and Pricing
GPT-5 is initially available to ChatGPT Pro subscribers and select API customers, with broader rollout expected over the coming weeks. Pricing for API access starts at $15 per million input tokens and $60 per million output tokens for the full-capability model, with a smaller, faster variant priced significantly lower for high-volume applications.
Enterprise customers will gain access through OpenAI’s business tiers, with custom pricing available for large-scale deployments. OpenAI says it expects the model to be available in the free ChatGPT tier by late 2025, though with capability limitations.
Industry Reaction
The launch has sent ripples through the AI industry. Shares of companies competing with OpenAI — including those backed by Google, Amazon, and independent startups — dipped in premarket trading following the announcement. Analysts noted that the performance gap between GPT-5 and competing models appears to have widened, at least based on early benchmarks.
Anthropic, whose Claude models have been strong competitors, declined to comment directly on GPT-5’s launch. Google DeepMind pointed to its own Gemini Ultra as remaining competitive on specific workloads.
Safety and Alignment
OpenAI says GPT-5 underwent more rigorous safety evaluation than any previous model, including extensive red-teaming by external researchers. The company published a 60-page system card detailing risks identified during testing and mitigations deployed in the final model. Notable improvements include better refusal of harmful requests without over-refusing legitimate queries — a balance OpenAI has struggled with in earlier models.
