Free models (opencode/*-free) silently throttled from datacenter IP since Oct 17 #56719

Description

Since Oct 17, all opencode/*-free models stopped responding from my cloud VM — first with rate-limit errors in the logs, then complete silence (requests hang, no output, no error surfaced).

Affected: opencode/deepseek-v4.1-flash-free, opencode/glm-5.3-flash-free, opencode/space-bunny-free, opencode/beam-free.

Steps to reproduce

1. On a Linux cloud VM with a fixed datacenter egress IP, sign in via opencode auth login
2. opencode run --model opencode/beam-free "reply with exactly: hi" (60s timeout)
3. Empty output, exit 0 — same for the other three free models

Ruled out

• Account: fresh OpenCode account on a second email → same silence on both. Reverted to primary.
• Client: upgraded 1.19.0 → 2.0.32; the throttle appeared after the upgrade, on the new client.
• Network basics: model catalog lists fine, opencode.ai loads fine (HTTP 200) from the VM.
• The same models respond intermittently from my phone on a different network.

Logs

~/.local/share/opencode/log/ shows AI.Error: Rate limit exceeded on Oct 17 ~08:58 IST; silence began ~13:00 IST Oct 17. Still silent as of Oct 18 13:30 IST.

Timing note

Failures began around the time of the disclosed credential incident. Not claiming a connection — flagging in case free-tier limits were tightened during incident response.

Ask

Is there IP-based rate limiting on free inference? If so, is it temporary, and is there an estimated reset? Happy to share timestamps and the egress IP privately. (Also emailed [email protected].)

Environment

• Version: 2.0.32
• OS: Ubuntu 24.04.5 LTS x86_64 (cloud VM)
• Plugins: none
• Terminal: non-interactive shell