AI Daily · Oct 25, 2026: OpenCode's new inference rack serves 6.77T tokens in a day
OpenCode's new DeepSeek inference rack handled 6.77T tokens in 24 hours at 1.16s p50 TTFT and 143 TPS; Command Code added DeepSeek V4.1 Flash and two more models plus a $10/month plan.
Pick a date
-
OpenCode reports 6.77T tokens served in 24 hours on new GPU rack
OpenCode's new DeepSeek inference rack handled 6.77T tokens in 24 hours with p50 TTFT of 1.16s and p50 throughput of 143 TPS; p95 is degraded by one faulty tray.
-
Command Code adds DeepSeek V4.1 Flash, MiniMax M3.1 Flash, Kimi K3.1 Pro
Command Code says DeepSeek V4.1 Flash 1017, MiniMax M3.1 Flash and Kimi K3.1 Pro are now live and available in its coding agent.
-
Command Code offers $10/month GOAT plan for its coding agent
Command Code's coding agent can be installed via npm and tried on a $10/month GOAT plan.
-
LiteLLM adds official support for OpenCode Go subscriptions
LiteLLM now routes OpenCode Go subscription models through existing setups, selecting the upstream API, preserving session state and recording estimated usage.
-
GMI Cloud MCP server exposes 36 read and 42 write tools by default
gmi mcp serve exposes 36 read tools by default, adds 42 write tools with --allow-write and 12 delete tools with --allow-destructive; sandbox tokens and API keys are never exposed and output is redacted and capped.
-
Neolab hedges on GPU capacity as H100/B200/B300 rental prices rise 2-3x
A neolab reportedly locked in GPU capacity early and can resell compute at 2-3x higher H100/B200/B300 rental prices if training fails; commenters note shortages are often followed by gluts.