Decode · tok/s
51
Prefill
—
tok/100W
72.9
Flash-Next Coder (ISTA GSQ-RCO) on llama.cpp. 51 tok/s writing, 168 tok/s on code edits. Mac Studio, M5 Max, 64 GB.
First-person M5 Max 64GB. Rank is the quoted 51 writing. 168 is code edits. Bit width was not named. 880 to 1,200 reading is a range, not a prefill rank.
