**Title:** [OpenCode Go] All Go models generate at ~2 tokens/s after updating to v2.0.32 #56733

Activity

1. opencode-agent commented on Oct 18, 2026

   

   opencode-agentBot

   on Oct 18, 2026 – with OpenCode Agent

   Contributor

   More actions

   Thanks for the report, and sorry Go has been so slow for you. Several parts of the issue still have template placeholders in them, so we can't yet tell whether OpenCode 2.0.32 is causing this or the Go service is slow. Could you fill in the following, in your own words?

   • Model(s): the exact Go model(s) you tried, as shown in /models (e.g. opencode-go/...)
   • Previous version: the last version where Go was fast for you, and how you installed OpenCode (npm, installer, scoop, etc.)
   • Rollback: does going back to that version make it fast again? You can test with npm i -g @opencode/cli@<version>. This is the most useful thing to know.
   • Other providers: on 2.0.32, is a non-Go provider (e.g. your own API key) also slow, or only Go?
   • Timings: for a simple prompt like "hello", roughly how long until the first token, and how long until the reply finishes?
   • Network: does this happen with no VPN or proxy?
   • Plugins: any plugins you have installed, or "none"
   • Logs: the log for a slow session, if you can get it. To make one, run opencode --print-logs --log-level DEBUG, reproduce the problem, and paste the relevant lines (remove any secrets first).

   You can reply here with the details or edit the issue. If you edit it, please also leave a comment saying you've updated it, so we know when you're done.

   These issues may be related. The list is ordered from most to least likely, so please check whether one of them already covers your problem:

   • Extremely Slow Response Time #56345: very slow responses (40 minutes for a simple reply), details still pending
   • Zen Go: deepseek-v4.1-flash decodes ~30% slower than the same model on Command Code and DeepSeek's own API #51010: Go model decoding slower than the same model on other providers
   • Zen Go: deepseek-v4.1-flash decode throughput down to ~119 tok/s median (254 at launch, 183 when reported in #51010) #55579: Go decode speed dropping over time
   • Slow after upgrade to v2 #52294: slowness after upgrading to v2 (closed)
   • Free models fail inside official OpenCode TUI with "free tier can only be used from within OpenCode" (Windows, v2.0.32) #56324 / [Bug] Free tier returns "can only be used from within OpenCode" inside OpenCode Desktop 2.0.32 — persists after restart and full reinstall of both CLI and Desktop; no login option in Desktop #56701: the separate "free tier can only be used from within OpenCode" error on 2.0.32
2. opencode-agent

   added

   triaging

   on Oct 18, 2026
3. Hemenguelbindi commented on Oct 18, 2026

   

   Hemenguelbindi

   on Oct 18, 2026

   Author

   More actions

   Thanks for the quick reply! Here's what I have:

   • Model(s):
     • MiMo-V2.6-Pro (Go) — works, but extremely slow: ~2 tokens/s
     • DeepSeek-v4.1-flash (Go) — doesn't respond at all, returns error 500
   • Previous version: unfortunately I don't remember the exact version — I accepted the in-app update prompt. It was a 1.x/pre-2.0.32 release where Go models ran at normal speed. Installed via <npm / installer / scoop>.
   • Rollback: <not tested yet / tested with @opencode/cli@<version> — result: ...>
   • Other providers: I don't have another provider API key to compare with, so I can't test a non-Go provider right now.
   • Timings: very long — <first token after ~N s, reply finished after ~N min> on MiMo-V2.6-Pro.
   • Network: <reproduced with no VPN or proxy / I use a VPN — will retest without it>
   • Plugins:
   • Logs: will attach a debug log from opencode --print-logs --log-level DEBUG <below / shortly>.

   The 500 error on DeepSeek-v4.1-flash may be related to #56717 and #56713, which were opened today with the same error on v2.
4. opencode-agent commented on Oct 18, 2026

   

   opencode-agentBot

   on Oct 18, 2026 – with OpenCode Agent

   Contributor

   More actions

   Thanks for the details, that helps.

   The 500 on DeepSeek V4.1 Flash doesn't look specific to your setup. Several reports opened today show the same error, and people commenting on them think it's a problem on the service side:

   • OpenCode Go: DeepSeek V4.1 Flash returns HTTP 500 while GLM works with the same credentials #56721: DeepSeek V4.1 Flash on OpenCode Go returns HTTP 500 while GLM works with the same credentials (closest match)
   • Error code 500 when using opencode x64 v2 #56713, Error 500 for all the models #56717, Provider request failed with HTTP 500 #56724: HTTP 500s reported more broadly since today

   Please follow those issues for the 500 rather than this one.

   The ~2 tokens/s on MiMo-V2.6-Pro may be caused by the same service trouble rather than by OpenCode 2.0.32. Could you test again once the 500s above are resolved? Then reply with:

   • Speed: whether MiMo-V2.6-Pro is still around 2 tokens/s
   • Timings: for a simple "hello", roughly how many seconds until the first token, and how long until the reply finishes. Your comment still has the placeholders there.
   • Install method: npm, the installer or scoop
   • Network: whether it happens with no VPN or proxy
   • Plugins: any plugins you use, or "none"
   • Log: the log from opencode --print-logs --log-level DEBUG for a slow reply, with any secrets removed

   About the previous version: if it was a 1.x release, that is a separate implementation from v2, so the comparison only helps if you know the exact version. If you can find it, testing the same prompt on it would show whether the client makes a difference.

   You can reply here or edit the issue. If you edit it, please also leave a comment so we know you've updated it.
5. opencode-agent commented on Oct 18, 2026

   

   opencode-agentBot

   on Oct 18, 2026 – with OpenCode Agent

   Contributor

   More actions

   Thanks for the report. We can't act on this as filed: the issue and follow-up still contain template placeholders, with no previous version, timings, rollback result or logs. If you still see unusually slow generation on a current build, please open a new issue with real timings for a simple prompt and a --print-logs --log-level DEBUG excerpt.
6. opencode-agent

   closed this as not plannedon Oct 18, 2026
7. opencode-agent commented on Oct 21, 2026

   

   opencode-agentBot

   on Oct 21, 2026 – with OpenCode Agent

   Contributor

   More actions

   We haven't heard back with the details we asked for, so we're closing this issue for now. We were waiting for a retest of MiMo-V2.6-Pro on OpenCode Go once the service-side HTTP 500 problems were resolved, along with timings and a debug log (opencode --print-logs --log-level DEBUG). Without those, we can't tell whether the slowness comes from OpenCode 2.0.32 or from the Go service.

   For the 500 on DeepSeek V4.1 Flash, please follow #56721 (and #56713 / #56717 / #56724).

   If Go models are still very slow for you now that the service has recovered, please reopen this or open a new issue with those details, and we'll look into it. Thanks!
8. opencodejr

   reopened this on Nov 3, 2026
9. neriousy

   closed this as not plannedon Nov 12, 2026