OpenCode says its own GPUs served 54% of its tokens, 13.4 trillion of them, and that it will have 10x the hardware in the next few months. The company behind the open-source coding agent posted the figures on X on 19 November. A token is the small chunk of text a model reads or writes.
The post doesn’t say what period “today” covers, and it gives no total. At 54%, 13.4 trillion tokens implies roughly 24.8 trillion across all of OpenCode’s traffic, but that’s arithmetic on the two figures, not a number OpenCode published.
54% of our tokens today were served by our own GPUs
— OpenCode (@opencode) November 19, 2026
that’s 13.4T tokens and we’re going to have 10x the hardware in the next few months
let’s go

OpenCode comes from the Anomaly team and sells model access two ways: Zen, which charges per request, and Go, a subscription from US$10 a month. Go’s docs say the plans work through “bulk discounts and reserved GPU capacity”.
Dax Raad, who builds OpenCode at Anomaly, gave the hardware detail two days earlier. He wrote on 17 November that “we now have 1000 GPUs”, 40% GB300s and 60% B300s in hand, “all on short term contracts”. GB300 and B300 are Nvidia data-centre GPUs.
i posted this a month ago and we now have 1000 GPUs
— dax (@thdxr) November 17, 2026
40% GB300s and 60% B300s in hand
all on short term contracts
everyone saying gpu shortage, impossible without 5 year commits, credit is dead blah blah
everyone just repeating what someone else told them

OpenCode’s CEO, Jay V, told YC’s Root Access in July that “we rent GPUs”, adding that OpenCode also works with providers that sell inference and with the model labs themselves. The 54% post doesn’t say what the other 46% ran on.
On 7 November Raad said OpenCode had landed a GB300 rack on a commitment of under a year, ready within a week. In a reply he said OpenCode serves 6 to 7 trillion tokens a day on that rack, with the B300s still in testing.
The post doesn’t say which models run on the GPUs. Raad’s 18 October post on a first test run named only a “new deepseek inference setup” and said its price and performance beat “every single vendor that we’ve found”.
OpenCode hasn’t said how many GPUs the 10x covers. Raad said on 27 October that he was “in the market for GB300 racks again” and “will be going with whoever can deliver soonest”.
Discover more from Sulat AI & Tech News
Subscribe to get the latest posts sent to your email.




