OpenAI, Thursday:

GPT‑5 is smarter across the board, providing more useful responses across math, science, finance, law, and more. It’s like having a team of experts on call for whatever you want to know.

The accompanying 75-minute introduction video includes several impressive demos, including several where GPT-5 wrote hundreds of lines of (apparently working!) code in just a few minutes.

(Sam Altman introduces the video, then—mercifully—steps off stage and leaves the presentation to his team. He continues to appear awkward and uncomfortable in front of an audience.)

This new model, they say, has “Expert intelligence for everyone.” It’s described as “safer and more accurate” and “less prone to hallucinations and pretending to know things.” This is especially important given their focus on health:

Our best model yet for health-related questions, providing more precise and reliable responses while acting as more of a proactive thought partner.

Health is apparently one of their most asked about topics. The video introduction included a segment where Altman interviewed a cancer survivor about her use of ChatGPT to help her understand her diagnosis. This is not an area for confabulations—that is, making shit up.

GPT-5 is already available to most customers. Curious what it had to say for itself, I prompted it, Tell me about GPT-5:

Alright, let’s cut the fluff—here’s the real deal on GPT‑5, dropped fresh on August 7, 2025.

The Rundown: What’s So Hot About GPT-5?

• Unified, smart-as-hell system: No more awkward model selection. GPT‑5 auto-routes queries—whether you need quick answers or deep thinking, it decides. Say “think hard about this” and it kicks into high-reasoning mode.

• Big gains everywhere:

• Writing, math, coding, health, visual perception—all of it cranks up performance across the board.
• Benchmarks: AIME 2025 math score hits 94.6%, SWE‑bench Verified coding at 74.9%, Aider Polyglot at 88%, multimodal understanding at 84.2%, and HealthBench Hard at 46.2%.

• Coding wizardry: Smarter front‑end generation, debugging, and “vibe coding” that feels like magic from a design-infused AI.

• Safer and sharper: Way fewer hallucinations, better at saying “I don’t know,” less goody‑two‑shoes yes‑man behavior, and real improvements in health advice.

• Context monster: Huge token window—270k input, 128k output (visible + invisible reasoning) for the API, and around 256–400k in other variants. Multimodal: text + image in, text only out.

• Pricing & variants:

• ChatGPT access: Free users get limited play; Plus gives more; Pro (~$200/month) brings unlimited and extended reasoning.
• API tiers: gpt‑5 (full), gpt‑5‑mini, and gpt‑5‑nano—pick your speed/cost combo.

Enterprise upgrades: Smarter Microsoft Copilot, better reasoning, deeper conversation tracking.

Bottom Line (Told Straight)

GPT-5 is basically a PhD-level AI roommate—sharp across the board, fast, and way more reliable than its predecessors. OpenAI clearly focused on being useful, not just flashy. It’s less “AI showing off” and more “AI that works for you.”

If you’re into writing clearer, reasoning deeper, coding smoother, or just want a smarter base-layer assistant—GPT-5 probably rules.

It includes links to recent stories from The Atlantic, The Washington Post, and Financial Times.

GPT-5 is certainly not afraid to toot its own horn.