[BUG]: OpenCode Zen free models hy3-free and nemotron-3-ultra-free failing while deepseek-v4-flash-free works #38028

Description

OpenCode Zen free models are inconsistently working as of v1.17.13. Direct API tests against https://opencode.ai/zen/v1/chat/completions show that two of the six listed free models fail with errors, while four succeed.

Failing models:

• hy3-free → {"code":30001,"message":"Sorry, your account balance is insufficient","data":null}
• nemotron-3-ultra-free → {"error":{"type":"server_error","message":"Error from provider (Console): Upstream request failed"}}

Working models (verified):

• deepseek-v4-flash-free → success, cost: 0
• big-pickle → success, cost: 0
• mimo-v2.5-free → success, cost: 0
• north-mini-code-free → success, cost: 0

This is not a client-side configuration issue — the failures occur when calling the Zen API directly with curl, with no authentication header required.

OpenCode version

1.17.13

Steps to reproduce

1. Run opencode --version (v1.17.13)
2. Test hy3-free:

curl -s https://opencode.ai/zen/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model":"hy3-free","messages":[{"role":"user","content":"say hi"}],"max_tokens":20,"stream":false}'

Expected: successful free response with cost 0.
Actual: {"code":30001,"message":"Sorry, your account balance is insufficient","data":null}

3. Test nemotron-3-ultra-free:

curl -s https://opencode.ai/zen/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model":"nemotron-3-ultra-free","messages":[{"role":"user","content":"say hi"}],"max_tokens":20,"stream":false}'

Expected: successful free response with cost 0.
Actual: {"error":{"type":"server_error","message":"Error from provider (Console): Upstream request failed"}}

4. For comparison, the following models work:

# deepseek-v4-flash-free - OK
curl -s https://opencode.ai/zen/v1/chat/completions -H "Content-Type: application/json" -d '{"model":"deepseek-v4-flash-free","messages":[{"role":"user","content":"say hi"}],"max_tokens":20,"stream":false}'

# mimo-v2.5-free - OK
curl -s https://opencode.ai/zen/v1/chat/completions -H "Content-Type: application/json" -d '{"model":"mimo-v2.5-free","messages":[{"role":"user","content":"say hi"}],"max_tokens":20,"stream":false}'

# north-mini-code-free - OK
curl -s https://opencode.ai/zen/v1/chat/completions -H "Content-Type: application/json" -d '{"model":"north-mini-code-free","messages":[{"role":"user","content":"say hi"}],"max_tokens":20,"stream":false}'

Screenshot and/or share link

N/A (API responses captured above)

Additional context

• The model list at GET https://opencode.ai/zen/v1/models still lists both hy3-free and nemotron-3-ultra-free as available models (owned_by: opencode), so they should be routable.
• This may be related to prior issue bug: "Insufficient Balance" error when executing free models (opencode/big-pickle) on OpenCode Zen #35149 (Jul 3, 2026) where all free models briefly showed "Insufficient Balance" due to a routing bug; that was fixed, but hy3-free appears to have a persistent failure now.
• nemotron-3-ultra-free may share root cause with NVIDIA NIM issues reported in Nvidia NIM API hangs for DeepSeek v4 reasoning models without chat_template_kwargs #24264 / Bug: NVIDIA API models (NIM) hang indefinitely during generation (Build/Thinking phase) despite successful model fetching #28385, though this is specifically the OpenCode Zen free tier, not self-hosted NVIDIA NIM.