ox-alpha · LiveCodeBench release_v6

Independent evaluation · greedy decoding · n=1 · sequential · generated 2026-08-21 20:48 UTC
28.0%
pass@1
49 / 175
problems solved
$0
inference cost
0
generation errors

Score by difficulty

easy
51.2%
22 / 43
medium
30.8%
16 / 52
hard
13.8%
11 / 80

Published context — LiveCodeBench

Modelpass@1source
DeepSeek V4-Pro93.5%
vendor-reported
DeepSeek V4-Flash (Think-Max)91.6%
vendor-reported
DeepSeek V4-Flash (Think-High)88.4%
vendor-reported
DeepSeek V4-Flash (Non-Think)55.2%
vendor-reported
ox-alpha (raw, measured here)28.0%
measured

Per-problem results

Reading this honestly. Vendor numbers come from each lab's own evaluation setup and windows; ours is an independent minimal evaluator on release_v6 (greedy, single attempt, subprocess timeout 20 s/test). Cross-evaluation deltas of a few points are normal — treat this as one measurement, not ground truth.