#kimi 2.5 cloud suddenly stopped working

1 messages · Page 1 of 1 (latest)

solid vapor
#

I was trying to start a new chat session, and openclaw reports the following error in my discord channel:
403 {"error":"this model requires a subscription, upgrade for access..."}
This seems more like a problem with ollama than openclaw, but I wanted to know if a) I accidentally broke something, and b) what local models I can use as a fallback in these kinds of situations.

prisma plazaBOT
rapid marlinBOT
#

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

glass perchBOT
#

Probably not, this smells upstream.

A 403 {"error":"this model requires a subscription..."} is usually Ollama Cloud refusing that specific :cloud model, not OpenClaw breaking itself. If it started suddenly, I’d suspect account/plan access or Ollama cloud availability before OpenClaw config.

First checks, in this order:

openclaw --version
openclaw status
openclaw models status
openclaw models list --provider ollama

If you’re using Ollama Cloud + Local, also make sure the Ollama host is still signed in:

# 
ollama signin

For fallback behavior, OpenClaw does support automatic model failover via agents.defaults.model.fallbacks, and billing/subscription-style failures are part of that failover path. So the usual setup is:

  • keep your preferred cloud model as primary
  • add a local Ollama model as fallback
  • let OpenClaw switch when the cloud model gets denied

Example command shape:

openclaw models fallbacks add ollama/<your-local-model>

For which local model to use, I would not hardcode a stale list here. Best path:

#
  • see what OpenClaw/Ollama can already see:
    openclaw models list --local
ollama list
  • then ask in #models for current real-world picks on your hardware

General rule from the docs: use the largest local model you can run reliably. Small/heavily quantized local models are much less reliable for agentic + tool-heavy use.

Docs:

If you paste the output of the first 3 commands, I can help narrow down whether this is just an Ollama subscription issue or a config mismatch.

solid vapor
#

I tried switching to deepseek-r1 but it returns the message
"LLM request failed: provider rejected the request schema or tool payload."

glass perchBOT
#

Yep, that looks like a different failure mode.

So right now you’re probably hitting two separate things:

  1. 403 ... requires a subscription
    → the Ollama cloud model is being denied upstream

  2. provider rejected the request schema or tool payload on deepseek-r1
    → the local model/backend is choking on OpenClaw’s tool schema

So no, this does not look like you “broke OpenClaw” so much as:

  • cloud access disappeared for one model
  • your local fallback is not happy with tool-heavy agent payloads

What I’d try next:

# 
openclaw --version
openclaw models status
openclaw models list --local

For the local fallback specifically, the docs point to two mitigations:

  • First try:
    agents.defaults.experimental.localModelLean: true
    This trims heavyweight default tools and makes local backends less brittle.

  • If it still fails:
    mark that local model with
    compat.supportsTools: false
    so OpenClaw stops sending full tool schema to it.

#

Important caveat: if you go the explicit models.providers.ollama.models[] route to set compat.supportsTools: false, that moves you out of Ollama auto-discovery and into manual model definitions, so I’d try lean mode first.

My honest recommendation: I would not use DeepSeek-R1 as the only fallback for full agent sessions unless you’ve tested it with tools. Local fallbacks are best treated as a safety net, and the “best” current local pick is moving fast enough that I’d check #models for fresh real-world recommendations on your hardware.

Docs:

If you paste openclaw --version plus your current model config block, I can suggest the smallest safe JSON5 change.

solid vapor
#

Oh I see the issue. Ollama had to make temporary changes to the kimi models, locking them behind paid subscription due to usage caps

glass perchBOT
#

Yep, that lines up.

So the 403 was upstream Ollama/Kimi access policy, not you breaking OpenClaw. Nice catch.

The deepseek-r1 error is still a separate local-model/tool-schema compatibility thing, but at least the original mystery is solved.

If you want to harden this a bit, I’d still set a tested fallback in agents.defaults.model.fallbacks so a cloud-model lockout doesn’t fully brick new sessions.

Glad you found it 🎉
If this fixed it, please right-click the solution message → Apps → ✅ Mark Solution so Answer Overflow can index it for the next person.