4 articles tagged
Open weights closed the intelligence gap from 13 points to 6, yet enterprise share fell to 11%. New August 2026 data shows identical weights scoring 22% or 37% on tool calling depending only on the endpoint.
State-of-the-art function-calling agents still solve under 50% of real tasks and stumble on retries. A 2026 guide to evaluating LLM tool-use reliability with pass^k, tau-bench, BFCL, and your own eval harness.
Microsoft, AWS and Google all publish Arm performance gains, and all three use a different baseline. Cobalt 200 is measured against Cobalt 100, Graviton4 against Graviton3, and Axion N4A against x86. Here is how to read
Gradient-boosted trees stopped setting the accuracy frontier on tabular data in 2026. But the benchmark that shows it also shows why XGBoost and LightGBM still belong in production, and the answer is not the one most
