BREAKING
OpenAI Software Tweak Halves Inference Cost
0
%
cost reduction
0
June found
Few Hundred GPUs for Logged-Out Traffic
0
%
compute low end
0
%
compute high end
Software Tweak vs Jalapeno Chip
Software Tweak
new
●
Applies to existing models
●
Mainly logged-out traffic
●
Paid tier unconfirmed
Jalapeno Chip
June 2026
●
Hardware ASIC with Broadcom
●
Aims ~50% per-token savings
Methods Undisclosed, Impact Unverified
AI NEWS BLITZ
OpenAI reportedly found software tweaks that cut inference costs by more than half.