DeepSeek rerun through Fireworks reportedly reproduces tester's results

On Oct. 28, tester @j_dekoninck wrote that rerunning DeepSeek through Fireworks reproduced their results, unlike an original run they described as significantly worse. After investigating with Fireworks, the tester found no issue explaining the difference. Fireworks participant @dzhulgakov said neither side could reproduce the issue anymore and suggested it was a statistical outlier, without establishing why the first run differed.