Rapid token/plan exhaustion with deepseek-v4-flash-vision-exp — two plans drained far faster than expected

Description

I subscribed to two paid plans and both were nearly exhausted very quickly, far
faster than expected based on the model's advertised cost. I'm using
opencode-go/deepseek-v4-flash-vision-exp and it drained both plans in a very
short time, even though I used it mainly for normal coding tasks (text only, no
images).

I was advised this model was low-token-cost, but the consumption was extremely
high. Both plans were almost fully consumed after very little usage.

Questions:

1. Is token counting/billing working correctly for this model?
2. Is the full context being re-sent and billed on every message, multiplying cost?
3. What is the recommended low-cost model for coding without vision?

Screenshots of usage/billing attached.

Plugins

None

OpenCode version

1.18.20

Steps to reproduce

1. Start a session with model opencode-go/deepseek-v4-flash-vision-exp.
2. Use it for normal coding tasks (mostly text, no images).
3. Keep the conversation going for a moderate amount of time.
4. Observe that token/plan consumption is extremely high.

Screenshot and/or share link

No response

Operating System

macOS 26.4.1 (Build 25E253), arm64 (Apple Silicon)

Terminal

No response