Every Top AI Model Scores 0% on ProgramBench
Meta, Stanford, and Harvard launch ProgramBench, a brutal new benchmark that asks AI to build software from scratch. GPT…
🏷️ AI benchmarks
39 articles about 'AI benchmarks'
Stanford HAI Finds AI Benchmarks Hitting Ceiling
Stanford's HAI 2025 AI Index reveals that leading AI models now saturate most major benchmarks, raising urgent questions…
Hugging Face Launches Agentic AI Leaderboard
Hugging Face debuts an open leaderboard for evaluating agentic AI systems, bringing transparency to one of AI's fastest-…
Google Unveils Gemini 2.5 Ultra With 1M Tokens
Google DeepMind launches Gemini 2.5 Ultra, its most powerful AI model yet, featuring a million-token context window and …
GPT-5.5 Instant Launches as ChatGPT Default Model
OpenAI releases GPT-5.5 Instant with 52.5% fewer hallucinations and major math gains, replacing GPT-5.3 Instant for all …
Claude 4 Shatters Graduate-Level AI Benchmarks
Anthropic's Claude 4 sets new records on GPQA and other graduate-level evaluations, outperforming GPT-4o and Gemini Ultr…
Grok 3.5 Beats GPT-5 on Math Reasoning Tests
Elon Musk's xAI releases Grok 3.5, which outperforms OpenAI's GPT-5 across major mathematical reasoning benchmarks.
Claude Opus 4 Sets New Bar for PhD-Level AI Reasoning
Anthropic's Claude Opus 4 achieves state-of-the-art results on GPQA Diamond, outperforming OpenAI and Google on PhD-leve…
GPT-5 Turbo Sets Records in Multi-Step Reasoning
OpenAI's GPT-5 Turbo achieves breakthrough scores on complex reasoning benchmarks, outpacing rivals by significant margi…
Do AI Benchmarks Actually Measure Intelligence?
A growing debate among researchers and industry leaders questions whether popular AI benchmarks reflect genuine intellig…
Mistral AI Launches Codestral 2.0, Rivals GPT-5
Mistral AI releases Codestral 2.0, a code-focused LLM matching or exceeding GPT-5 on key coding benchmarks.
After Million-Token Context, What's the Next LLM Battleground?
The context window race is effectively over. The real competition now shifts to reasoning depth, efficiency, and archite…