Developer cancels $200 Codex subscription for local GLM-5.3-Flash

A developer says running GLM-5.3-Flash locally gives results close to GPT-6 Astra in their specific niche at a cost of about $0.06 per hour.

Iridescence

@iridescence_devArchived 2026.09.19Originally posted 2026.09.18

Iridescence @iridescence_devOriginally posted 2026.09.18

Open original on X

I feel vindicated in my choice to cancel my $200 Codex subscription and use @MiaAI_lab's GLM-5.3-Flash for local inference instead. In my specific niche it's giving me results comparable to GPT 6 Astra but at a cool $0.06 / hr.

x.com/iridescence_dev/status/2100974258406965489Archived 2026.09.19 10:18

Open original on X