We just rolled out CursorBench 4.0! It includes new tasks for how well models follow instructions, work on challenging projects over time, and is more difficult than before (so all models score lower).↗

Wrote up my initial impressions of Claude Fable 5 - it has a big model smell: slow, expensive and capable of crunching through pretty much everything I threw at it simonwillison.net/2026/Jun/9/cla…↗

I think 200k context window w/ good compaction is the sweet spot. Models claiming to have larger are rarely better at rembering important stuff from my experience↗