GPT-5 Benchmarks 42% on HLE and 75% on SWE Bench
LLMS
-
GPT-5 Mini Reaches O3 Level Performance for Free Users
By
–
GPT-5 mini on o3 level, when free user hit rate limit
-
GPT-5 Release: Major Advances in Multimodal AI and Code Generation
By
–
GPT-5 just dropped and the benchmarks look insane It is incredible at single-shot understanding of complex prompt, multimodal reasoning and writing code that just works. Huge leap from previous generation of OpenAI, Claude and Qwen AI models.
-
Introducing Align Evals: New Evaluation Framework for LLMs
By
–
Get started by reading our blog post: https://
blog.langchain.com/introducing-al
ign-evals/
…
Or, watch a video on how it works: -
LangSmith Align Evals for Consistent Reliable Evaluation Scores
By
–
Want more consistent, reliable evaluation scores? Last week we launched Align Evals in LangSmith — helping you craft better prompts to reduce inconsistencies, and build more reliable evals More details in the reply below
-
Secret access to GPT-5: all its capabilities revealed
By
–
OpenAI secretly gave me access to GPT-5 for weeks and I've tested it thoroughly for you. Here is everything it can do:
-
GPT-5 Demonstrates Superior Writing and Research Insights
By
–
GPT-5 is a very good writer, able to connect analogies between things you wouldn't expect older GPT series models to do. There are actual insights now in what it produces over generic word salad. Deep research also started returning niche numbers/insights that I didn't expect. 5
