Legacy-Bench: leaderboard

Coding agents fix bugs, add features and migrate code in COBOL, Java 7, BASIC, C89, Fortran and Assembly, each task checked by hidden tests; the score is the pass rate over 3 Fireworks runs of 100 tasks. SWE-bench-style boards use modern Python and JavaScript repositories instead.

Metric: Tasks passed, mean of 3 runs of 100 tasks (%). Source: fireworks.ai. Saturation forecast: Around 2031. 13 models tracked. Updated 16 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1Kimi K3.1 ProMax63#29
2Muse Spark 1.3Max62#43
3Claude Opus 5.5Medium62#5
4DeepSeek V4.1 FlashMax57.67#62
5DeepSeek V4.1 Flash (1017)Max56.33#41
6GPT-6 AstraMax55.67#6
7GPT-6.1 SolMax55#10
8Grok 4.7Max54#47
9Kimi K3Max52.67#56
10Gemini 3.8 Flash52.33#36
11GLM-5.3Max50.67#68
12Muse Spark OpenMax50#81
13Gemini 4 ArgonHigh44.33#12

No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon.

Interactive version: theaggregate.ai/benchmark?slug=legacy-bench · How It Works · Data refreshed daily, snapshot 2026-11-16.