#Glm 4.6 error

1 messages ¡ Page 1 of 1 (latest)

normal terrace
#

When I try to use this model I keep getying this error: Maybe a api error? ⚠️ Agent failed before reply: All models failed (1): zai/glm-4.6: Provider zai is in cooldown (all profiles unavailable) (rate_limit).
Logs: openclaw logs --follow

storm geyserBOT
#

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

timber boughBOT
# normal terrace When I try to use this model I keep getying this error: Maybe a api error? ⚠️ Ag...

That error is OpenClaw telling you it’s backing off because Z.AI rate-limited you:

  • Provider zai is in cooldown … (rate_limit) = the Z.AI/GLM API returned a rate-limit style failure (usually HTTP 429), so OpenClaw put all Z.AI auth profiles into a temporary cooldown and won’t send more requests until it expires.
  • Because you only tried one model (zai/glm-4.6) and have no usable Z.AI profile right now, you get All models failed.

What to do:

  1. Wait a bit and retry (cooldowns are usually minutes). Restarting the gateway typically won’t “fix” rate limits.

  2. Confirm what OpenClaw thinks is happening:

openclaw models status --json

Look for auth.unusableProfiles entries for provider zai (it should show the reason/cooldown state). If you paste just that section here, we can tell if it’s a short rate-limit cooldown vs something more like billing/credits.

  1. Check Z.AI-side limits/credits in the Z.AI console for your ZAI_API_KEY (some keys/plans have very tight QPS / concurrent-request limits).
#
  1. Add a fallback model/provider (so you still get replies while Z.AI is cooling down). Docs on failover: https://docs.openclaw.ai/help/faq#model-failover-and-all-models-failed

One extra gotcha: if you ran openclaw models status --probe, that does real API calls and can itself trigger rate limits. So avoid probing repeatedly while debugging.

If you can paste:

  • openclaw --version
  • openclaw models status --json → the auth.unusableProfiles bit
  • and the log lines right before the cooldown message (from openclaw logs --follow)
    …I can be more specific about whether it’s quota, concurrency, or a transient 429.
normal terrace
#

I did not manage to use this model for a few days and tried daily I think they changed their api format

quick temple
#

just configured to GLM .. no problem

quick temple
limber ermineBOT
# quick temple https://x.com/thatsnas1oo/status/2020965637535777205?s=46

@thatsnas1oo via Twitter

Gᙇ (@thatsnas1oo)

so easy switching from kimi to glm coding ✅
︀︀
︀︀kimi was solid for experimenting ~ definitely gets the job done. once i started pushing past small tasks, i hit the wall (52-hour reset).
︀︀
︀︀been hearing about glm for a minute, so i’m tapping in now to see what i can really build 🚀
︀︀
︀︀full support for claude code, cline, plus 10+ top coding tools ~ starting at $3/month.
︀︀
︀︀solid option if u’re building, testing, or pushing agents without burning credits. limited-time deal live now.
︀︀
︀︀tap in here → z.ai/subscribe?ic=GQYRVWJFGM

**❤️ 1 👁️ 15 **

▶ Play video
FxTwitter ¡
quick temple
#

not a expert but my kimi plan has to wait 52 hours to reset 🤣🤣🤣

timber boughBOT
# normal terrace I did not manage to use this model for a few days and tried daily I think they c...

If Z.AI actually changed their API format, you’d typically see a 400/404 “invalid request” style error in the logs. The message you pasted is one layer above that: OpenClaw has already marked the zai provider as temporarily unusable and is refusing to even try it right now (“cooldown”).

Two important details:

  • Cooldowns are triggered not only by “real” 429s, but also by some auth/timeout/invalid-request failures (they’re treated as “failover-worthy”), so an upstream API change could still end up showing as “cooldown” after the first failure.
  • To know what’s really happening we need the original error that caused the cooldown, not the cooldown line itself.

Can you paste these (they’re safe; no API keys should be shown):

  1. Your version:
openclaw --version
  1. The provider/profile health (look for auth.unusableProfiles for zai):
# 
openclaw models status --json
  1. A log snippet from the first failure (scroll a bit above the “provider zai is in cooldown” line in openclaw logs --follow and paste the first zai/HTTP error you see — that’s usually where it says 429 vs 401 vs 400, etc.)

Once we see whether it’s 429 rate limit, insufficient credits/billing, invalid API key (401), or invalid request (400), we’ll know whether the fix is “wait”, “add credits”, “re-key”, or “update OpenClaw / provider adapter”.

#

If other folks can hit zai/glm-4.6 fine (like Nas just said), it’s very unlikely Z.AI “changed the API format” in a way that only breaks you. This looks much more like your Z.AI account/key is being rate-limited or quota/credits-limited.

A key detail: OpenClaw’s “cooldown” is just a backoff mechanism after a failure — it’s usually minutes (exponential up to ~1h). So if it’s been failing for days, that usually means every time you try, Z.AI is still returning a rate limit / quota error again, and OpenClaw backs off again.

Can you paste:

  1. The actual upstream error line (a few lines above the cooldown message) from:
openclaw logs --follow

We need to see if it’s HTTP 429 (rate limit), 401 (bad key), or a billing/credits message.

  1. This (look for auth.unusableProfiles for zai):
openclaw models status --json
#

Also double-check you’ve got your key set the supported way:

Once you paste the log + models status --json, we can tell you exactly whether it’s “wait for reset” vs “add credits” vs “key invalid” vs “update OpenClaw”.

normal terrace
#

Thanks