FreeUsageLimitError persists 10+ days on deepseek-v4-flash-free even for minimal requests — while other free models work

Description

The deepseek-v4-flash-free model on OpenCode Zen consistently returns 429 FreeUsageLimitError for any request from my client, including a minimal one-message request. The error has persisted for 10+ days with no recovery.

Error returned by the provider:

429 FreeUsageLimitError: "Rate limit exceeded. Please try again later."

This no longer looks like a usage-volume issue — even a single tiny request fails immediately.

Plugins

none

OpenCode version

N/A (using the OpenCode Zen API endpoint https://opencode.ai/zen/v1 via a self-hosted AI gateway)

Steps to reproduce

1. Send a request to https://opencode.ai/zen/v1/chat/completions.
2. Use model deepseek-v4-flash-free.
3. Send even a minimal request:

curl -X POST "https://opencode.ai/zen/v1/chat/completions" \
  -H "Content-Type: application/json" \
  -d '{"model":"deepseek-v4-flash-free","messages":[{"role":"user","content":"hi"}],"max_tokens":1}'

Result: HTTP 429 with FreeUsageLimitError.

Observed behavior

• deepseek-v4-flash-free → 429 (even a single "hi")
• mimo-v2.5-free → 429
• big-pickle → 429
• longcat-2.0-free → 429
• nemotron-3-ultra-free → 200
• laguna-s-2.1-free → 200

The failure is model-specific from the same client/IP, and has not cleared after 10+ days.

Expected behavior

• A minimal request should succeed, or if throttled, the limit should recover after the relevant window rather than staying blocked for 10+ days.
• At minimum, it should be possible to distinguish a temporary usage limit from a persistent per-IP/per-model flag.

Screenshot and/or share link

No response

Operating System

Linux (VPS)

Terminal

Self-hosted AI gateway (direct curl reproduces the 429 upstream)

Additional context

• Appears related to #42385, #33318, #33495, #32971.
• The model-specific pattern (some free models OK, others 429 from the same client/IP) suggests a per-model free-usage flag rather than a total block.