opencode now serves 44% of its 16T daily tokens on its own GPUs
AccBalanced · x · 2026-11-19
opencode reflects on building its own inference stack: DeepSeek 4.1's success and price hike in August destroyed their economics, and no inference provider could meet their needs. Despite owning zero hardware and having no GPU ops experience, three months later they're serving DeepSeek and GLM at the same quality as the labs themselves — 44% of today's tokens (16T) ran on their own GPUs, with 10x more hardware coming online in the coming months.
More from coding & agent
- Overmind launches autonomous AI lab: fine-tune your own model from production agent traces — jasonkneen · 2026-11-20
- One setting boosted OpenClaw's score 44%: feed reasoning content back to the model — TheZachMueller · 2026-11-20
- Dev raves about Context Engineering book: 'evals remain the critical piece humans must do' — iamrobotbear · 2026-11-20
- Anthropic publishes architecture guide for always-on, event-driven agents — rseroter · 2026-11-20
- One header bypasses vLLM Semantic Router guard: duplicate Istio filter name to blame — colinmcnamara · 2026-11-20
- What Sneakily Spiked Your LLM Bill? Cache-Killing Prompts and Billed Retries — Junior-Guitar-290 · 2026-11-19