Qwen3.8 Flash-Next on an RTX PRO 6000: the missing tests
On a 96 GB RTX PRO 6000, I compared three Qwen3.8 Flash-Next quantizations with Basalt. Q2_0 was fastest and all passed the hard canary; Q4 fit comfortably but took longer on the coding tests.
On a 96 GB RTX PRO 6000, I compared three Qwen3.8 Flash-Next quantizations with Basalt. Q2_0 was fastest and all passed the hard canary; Q4 fit comfortably but took longer on the coding tests.