Token counter broken in Desktop App with Synthetic provider (cost stuck at $0, usage at 0%, tokens frozen) #39092

Version

OpenCode Desktop v1.18.4, Windows 10, active Synthetic API key (syn_...)

Context

Using the OpenCode Desktop App with the Synthetic provider (both the built-in provider and a custom OpenAI-compatible provider).

This bug is related to two issues reported earlier today:

• Crash on startup: "Notification server not found: wsl:Ubuntu-24.04" after removing WSL server #39071 — Crash on startup after removing a WSL server
• Built-in Synthetic provider: "Authorization header required" + ghost models persist after disabling #39072 — Built-in Synthetic provider: "Authorization header required" + ghost models persist after disabling

All three issues share the same context: a state reset performed to recover from the WSL crash (#39071). The token counter bug appeared alongside the provider regression in #39072.

Problem

The token counter at the bottom of the session shows inconsistent values:

• Cost : stays at $0 regardless of conversation length
• Usage : stays at 0%
• Tokens : displays 28690 (value does not appear to evolve as expected, need to verify whether tokens decrease over the session)

Steps to reproduce

1. Launch the OpenCode Desktop App
2. Start a session with the Synthetic provider (built-in or custom)
3. Exchange several messages
4. Observe the counter at the bottom: cost at $0, usage at 0%, tokens frozen

Environment

• OpenCode Desktop v1.18.4
• Windows 10
• Provider: Synthetic (built-in) and Synthetic Custom (OpenAI-compatible, baseURL https://api.synthetic.new/openai/v1)
• The bug reproduces with both provider configurations

Hypotheses

• The Synthetic / custom provider may not return token usage metadata in its response, or OpenCode does not parse it correctly for this type of provider.
• The $0 cost may stem from a missing price/token entry in the models.dev catalog for Synthetic models.

Suggestions

1. Verify that token usage metadata parsing works for custom OpenAI-compatible providers.
2. Allow users to manually set the price/token for custom providers so cost can be computed.
3. When usage data is unavailable, display n/a instead of 0 to distinguish "no data" from "zero consumption".