GPT-5 dropped and devs are posting initial thoughts on HackerNews
OpenAI launched GPT-5 this week and the developer community on Hacker News had some interesting reactions. Here’s what people are actually saying about it.
Pricing & Performance
GPT-5 pricing came in at $1.25/$10 per million input/output tokens, with $0.125 for cached tokens. That’s significantly cheaper than Claude Opus 4.1’s $15/$75 pricing.
Performance-wise, it hit 74.9% on SWE-bench compared to Claude’s 74.5%. Pretty close, but at a much lower cost.
Developer Sentiment
The discussion showed mixed reactions. People appreciated the product-focused approach and live coding demos. Several commenters liked seeing actual features instead of just research papers.
But there’s definitely skepticism. Multiple developers pointed out that while the presentation looked impressive, the actual benchmark improvements were only around 5%. Some called out OpenAI for making their charts more dramatic than the data supported.
A common theme was that this feels like incremental progress rather than the major leap the version number suggests. People are calling it “diminished returns” and questioning whether it deserves the GPT-5 label.
Technical Reactions
Developers noted issues with the API rollout – some hitting verification loops when trying to access it. The gradual rollout to different user tiers also frustrated people wanting immediate access.
There were comparisons to Claude Code and existing AI coding tools, with questions about whether GPT-5 actually moves the needle much beyond what’s already available.
The Waiting Game
A lot of the discussion centered around waiting for real-world testing. People want to see how it performs on actual projects, not just polished demos. The general attitude seems to be cautious optimism mixed with healthy skepticism.
Several developers mentioned they’ll test it out, especially given the better pricing, but aren’t expecting it to fundamentally change their workflows.
Bottom Line
The HN crowd sees GPT-5 as a solid incremental update with competitive pricing, but not the revolutionary breakthrough some were hoping for. Most developers are taking a wait-and-see approach to judge its real-world performance beyond the launch hype.
My Take
GPT-5 seems like a solid upgrade with better pricing. Nothing groundbreaking, but that’s fine. Most of us just want tools that work reliably and don’t cost a fortune.
I’ll probably test it out for some WordPress projects, especially with those lower token costs. But I’m not holding my breath for it to revolutionize how I work in the short term.
AI Performance Curve
it reminds me of an awesome infographic created by the New York Times in the past, showing olympic running records over the past 120 years.
Credit: Screenshot of the New York Times Infographic
There is a definite curve there, and I think we may be in the midst of that long median curve with our AI models, at least for the foreseeable future.