fix(models): block unavailable Zen free models (deepseek-v4-flash-free, laguna-s-2.1-free)

[Fahad090NP]

Summary

Blocks two broken OpenCode Zen free models from the picker by adding them to KNOWN_UNAVAILABLE_MODEL_IDS. Both are listed by the gateway but fail every request:

Model	Error	Cause
deepseek-v4-flash-free	HTTP 400 Model is unavailable	Upstream model removed/broken (#203, #204)
laguna-s-2.1-free	HTTP 503 Endpoint is unavailable	Upstream endpoint down (transient 5xx retry exhausts)

On v0.7.2 the 400 also triggered a secondary crash (Body is unusable: Body has already been read, #199) — already fixed in v0.7.3, but the model itself remains broken.

Changes

• src/config.ts — added deepseek-v4-flash-free and laguna-s-2.1-free to KNOWN_UNAVAILABLE_MODEL_IDS (filtered on every path: live, cached, bundled)
• README.md — removed both models from the Zen free-models table
• CHANGELOG.md — added/extended the [Unreleased] entry

Closes

• Closes [BUG] Recursively Occurring with Deepseek v4 Flash (Free) as I Tested #203
• Closes [BUG] Recursively Occurring with Deepseek v4 Flash (Free) as I Tested #204

Fi Amanillah

[ltmoerdani]

hi @Fahad090NP, thanks for the thorough report and the quick turnaround on this, the error table in the PR description makes it really easy to follow.

The deepseek-v4-flash-free part looks good to me. Adding it to KNOWN_UNAVAILABLE_MODEL_IDS matches what we found when we looked at it during #182, the gateway lists it but it's broken upstream, so hiding it from the picker is the right call. Nice catch on the stale package-lock.json bump too (0.6.0 → 0.7.3).

I have a few small things on the laguna-s-2.1-free half before we merge:

1. The commit doesn't actually block laguna. Commit 4f42d5e9 says "block unavailable laguna-s-2.1-free from picker" but it only touches CHANGELOG.md, src/config.ts isn't changed, so laguna is still missing from KNOWN_UNAVAILABLE_MODEL_IDS and still shows up in the picker. The changelog entry says "both added", which isn't quite true right now. Could you either add it to the set in src/config.ts, or drop the laguna claim from the changelog/commit so the description matches the code?

2. Can you confirm laguna is consistently down, not just flaky? Context: laguna was actually verified working when we fixed [BUG] Deepseek does not show free models. #182 (we restored it from the deprecated filter because the gateway cross-check confirmed it was live — see docs/issues/78-20260822-issue182-deprecated-model-gateway-crosscheck.md). A 503 "Endpoint is unavailable" can also be a temporary upstream outage, which is how we treat qwen3.6-plus-free, instead of hard-blocking it, we surface a note via CAPACITY_LIMITED_MODEL_NOTES so users know to retry. If you've been seeing 503s from laguna for a while (multiple days / every attempt), then blocking is fair, just let me know how long it's been failing and I'm good with adding it. If it's more recent, a capacity note might be the safer option until it recovers.

3. Small doc nit: the laguna row is removed from the README table, but the prose above it still lists "Laguna S 2.1 Free" as currently free (same in the later section around line 460). Worth syncing those whichever way we go on point 2.

The deepseek-v4-flash-free fix alone is mergeable as-is, so if you'd rather split it, that works too. Let me know how you want to handle laguna and I'll take another look. 🙏

[Fahad090NP]

@ltmoerdani if your AI agent can do the desired job, its okay, otherwise close this PR with my issues. I had deleted this Project from my side and I do not want to work on this extension anymore, because I found DeepSeek Harness and it supports OpenCode API and works best for me. The Output, UX and even UI of Deepseek Harness is best and working absolutely fine for me. In VS Code, I were recursively facing multiple errors, so now I will be using Deepseek Harness. I also want you to try it, it works best and natively supports OpenCode's API. Thanks a lot for your amazing suggestions and way of saying. Thanks again for your good morals.

Fi Amanillah.

[ltmoerdani]

Thanks a lot for all the reports and PRs, @Fahad090NP your contributions to this project have been really appreciated. Hope DeepSeek Harness works well for your workflow. 🙏

Good news on the bug: before closing this PR, I tested laguna-s-2.1-free against the Zen gateway directly and it's back online (HTTP 200, normal response). The 503 was a temporary upstream outage, so:

• I'll keep laguna-s-2.1-free in the list (verified live) and revert its removal.
• deepseek-v4-flash-free stays blocked since it's confirmed broken upstream, I'll implement that part directly on main.

Thanks again, and good luck with your future projects. May Allah always bless you with beneficial knowledge.