As of today, none of our characters that use Llama3-70B are able to fire simple actions. Yesterday it worked fine and over the last few weeks we’ve actually found Llama3-70B to be one of the better action LLMs. Did something happen on the server side? All other characters are working that use other LLMs so the issue seems to be with Meta LLMs at the moment.
This broke many of our production characters and is causing big headaches for us.
Could you please share the session ID(s) where you experienced this issue?
It seems to be any character set to Llama3-70B
da79f269eb0d7fa63df57726c226a071
This is a known issue with the Llama3-70B model. The model that previously worked best with Actions has been deprecated, so actions are currently unreliable with Llama.
For now, we recommend switching your characters to a different LLM if you rely on Actions.
Thank you. This model used to be one of the most reliable in firing actions in our testing so it was unfortunate to see it break like it did and leave our product unusable. We may have to setup our own internal testing to ensure these services are active and monitor them if they go down.
Is there a notification we can reference for models that are having issues?
As you know, over time some models get deprecated by providers.
For now, my recommendation is to try other models. Gemini 2.5 Flash, Sonnet 4.5, GPT 4.1 are good options to explore. The Grok 4 is also not bad.
That’s fine, it’s just that we as a customer need a way to know this information if possible so that we can make appropriate changes to avoid down time.