SciCode leaderboard
SciCode
3 models tested · Updated 2026-06-19 · Verified sources only
Sakana Fugu leads at 60.1%
1
Sakana Fugu
Sakana AI · arxiv/2606.21228 · 2026-06-19
Single-worker routing. BEATS Fugu-Ultra (58.7) on SciCode.
60.1%
2
Sakana Fugu-Ultra
Sakana AI · arxiv/2606.21228 · 2026-06-19
Orchestrator result on SciCode.
58.7%
3
Muse Spark 1.1
Meta · Blog/Meta · 2026-07-09
Beats Gemini 3.6 Flash (52.7%). Strong scientific coding performance.
58.2%
benchmark.space canonical public ledger of AI capability