OpenAI Launches GPT-5 With Record Reasoning Scores

OpenAI Unveils GPT-5 as New Flagship Model

OpenAI has officially introduced GPT-5, positioning it as its most capable model yet across reasoning, coding, multimodal understanding, and specialized domains like health. The company says the model establishes new state-of-the-art benchmarks in several categories, reflecting a broader push across the industry toward AI systems that reason more deliberately rather than simply generating fast responses 12.

Benchmark Gains Across the Board

According to OpenAI, GPT-5 scored 94.6% on the AIME 2025 math competition without the use of external tools, a notable jump in mathematical reasoning performance. On coding tasks, the model reached 74.9% on SWE-bench Verified and 88% on Aider Polyglot, benchmarks designed to test real-world software engineering ability rather than isolated puzzle-solving. In multimodal understanding, GPT-5 posted an 84.2% score on MMMU, and in medical reasoning it achieved 46.2% on HealthBench Hard, a benchmark built to stress-test AI performance on difficult healthcare-related questions 1.

OpenAI also highlighted a specialized configuration called GPT-5 Pro, which uses extended reasoning time to push performance even further, scoring 88.4% on GPQA, a graduate-level science reasoning benchmark, without relying on external tools 1. The company frames these gains not as narrow lab results but as improvements that translate into tangible benefits for everyday users interacting with the model in practical settings 1.

Multimodal Reasoning and Broader Industry Context

Separate coverage of the launch emphasizes GPT-5's multimodal reasoning capabilities, describing its ability to integrate text, images, and video into a unified framework for understanding varied types of input 2. This reporting characterizes the upgrade as a roughly 40% improvement in complex problem-solving compared to GPT-4, suggesting implications for fields ranging from scientific research to software development and creative work 2.

While OpenAI's own presentation focuses on granular, benchmark-by-benchmark performance figures, outside coverage situates GPT-5 within a wider competitive landscape, describing the release as part of an intensifying race among AI developers to build more sophisticated reasoning models in 2025 2. Taken together, the two threads of coverage — one benchmark-driven, one industry-framed — point to the same underlying narrative: GPT-5 represents a meaningful step up in capability, particularly in tasks requiring multi-step reasoning, technical problem-solving, and cross-format understanding.

Why It Matters

The launch matters because it signals where large AI labs are directing their competitive energy: not just larger models, but ones optimized for sustained reasoning, tool-free problem solving, and multimodal comprehension. If the reported benchmark gains hold up in independent testing, GPT-5 could accelerate adoption in coding, research, and healthcare-adjacent applications, while intensifying pressure on rival labs racing to match or exceed its reasoning performance.