SWE-bench Multimodal Leaderboard Debuts, Claude Opus 5.5 Leads at 70.21%

A Princeton team led by Ofir Press open-sourced SWE-bench Multimodal, a harder benchmark focused on visual software bugs such as map rendering errors, and Vals AI launched its leaderboard. Claude Opus 5.5 tops the board at 70.21%.

2026-10-15 ~ 2026-10-15 · 4 related posts

• New SWE-bench Multimodal leaderboard: Claude Opus 5.5 leads at 70.21% on visual bugs — OfirPress · 2026-10-15
• ValsAI launches SWE-bench Multimodal leaderboard focused on visual software bugs — OfirPress · 2026-10-15
• Ofir Press's team open-sources a multimodal coding benchmark probing the capability frontier — OfirPress · 2026-10-15

1 near-duplicate retellings: OfirPress