Billing dispute — unintended fallback consumed ~25% of my OpenCode Go limit #46894

Description

Hello OpenCode Support,

I’m contacting you regarding an unexpected and significant usage of my OpenCode Go allowance.

A hidden/automatic fallback mechanism in the OMO beta / Senpi setup switched to expensive models without my explicit knowledge or consent. As a result, approximately 25% of my Go usage limit was consumed.

The main issue is that I did not intentionally select or authorize those expensive models for this usage. The fallback mechanism made the model selection automatically, and the consumed tokens did not result in a useful solution to my tasks or problems.

I therefore believe this usage should be reviewed as unintended/incorrect usage rather than normal user-initiated model consumption.

I would like to request:

A review of the relevant usage logs and model selections.
Confirmation of which models were used by the OMO beta / Senpi fallback and how much of my Go allowance they consumed.
Restoration/refund of the approximately 25% of my Go allowance that was consumed by this unintended fallback behavior.
Clarification of whether this fallback behavior is expected, and if so, why users are not explicitly warned before switching to significantly more expensive models.

I understand that OpenCode Go has different usage multipliers depending on the model. However, that is precisely why I believe an automatic fallback to expensive models without clear user consent deserves a billing adjustment.

I can provide screenshots, timestamps, logs, and other information needed to identify the affected requests.
Why it was not my choice.
My omo/opencode whitelist allows only free + budget models (ling-3.0-flash-fin-free, deepseek-v4-flash, hy3, qwen3.7-plus)

And I only just found out that model fallback is enabled for me; now I’ve already managed to find it and have probably disabled this function.Idk.

I set model_fallback: false. Nothing paid was selected or enabled for me. The agent nonetheless auto-routed my requests to these paid models.

Evidence (attached: fallback.log):

Kimi K3 (2026-08-30, repeated):

• opencode/deepseek-v4-flash -> opencode/kimi-k3 (billing), then opencode/kimi-k3 -> opencode-go/kimi-k3 (billing) - actual paid Kimi K3 execution.
• opencode/hy3-free -> opencode/kimi-k3 (hard-error), opencode/ling-3.0-flash-fin-free -> opencode/kimi-k3 (transient).

DeepSeek (2026-08-29 and 2026-08-30):

• opencode/claude-sonnet-5 -> opencode-go/deepseek-v4-pro (billing) on 2026-08-29 - paid DeepSeek V4 Pro auto-selected as fallback.
• opencode/deepseek-v4-flash repeatedly hit CreditsError (401, insufficient balance) on 2026-08-30, i.e. this paid model was charged, then auto-routed to Kimi K3.
• On 2026-08-29, a billing failure cascaded through multiple paid models instead of stopping (05:43 UTC), triggering additional paid attempts.
• Candidate churn: candidate_skipped kimi-coding/k3:max 46x.

Please treat this as a billing/usage dispute and escalate it to the appropriate team if necessary.

Thank you.

fallback.log
fallback_evidence_raw.txt

Plugins

omo beta \ omo senpi

OpenCode version

omo (omo-ai beta 5.0.0-0.beta.31)

Steps to reproduce

No response

Screenshot and/or share link

No response

Operating System

Windows 10

Terminal

wave