Cerebras has retired two models from its public catalog. As of May 27, 2026, requests to the public Llama 3.1 8B and Qwen 3 235B Instruct endpoints are no longer served. Private and dedicated deployments aren't affected, and Cerebras still offers GPT-OSS and GLM 4.7 on its pay-as-you-go tier — but the two public endpoints are gone.
- •Llama 3.1 8B — public endpoint retired
- •Qwen 3 235B Instruct — public endpoint retired
If your code called those Cerebras endpoints directly, you now have a migration on your hands: find a new provider, sign up, swap base URLs and model strings, re-test. Llama 3.1 8B in particular is still part of plenty of active production workflows, and "this endpoint stops answering on a date" is exactly the kind of change that turns into an incident.
One provider retiring a model should not be your problem
This is the same story as every model deprecation. A model you pin to a single provider's endpoint can be retired on a schedule, rate-limited, re-priced, or moved behind a paid tier — and when it is, your app is the one that breaks. The fix isn't to chase the model to whichever provider still has it this month. It's to stop depending on any one provider for it at all.
On Antbase, these models were never tied to Cerebras
On Antbase, a model string like meta-llama/llama-3.1-8b-instruct or alibaba/qwen3-235b-a22b isn't a single upstream endpoint — it's a routing target. ANT sends each request to whichever provider is currently healthy and cheapest for that model, and falls back automatically if one is down. Both of these models are carried by several providers in the catalog: Llama 3.1 8B is served by Featherless, Novita, and Hugging Face; Qwen3 235B by Novita, Scaleway, NetMind, Chutes, and more. Cerebras was only ever one option in that pool — so its public endpoints going away changes nothing about whether you can call these models on Antbase.
Same model string, same OpenAI-compatible call. Nothing to migrate:
from openai import OpenAI
client = OpenAI(base_url="https://antbase.ai/v1", api_key="ant-...")
# Routes across every provider that still serves these models,
# with automatic fallback. Cerebras retiring its endpoint is a no-op here.
resp = client.chat.completions.create(
model="meta-llama/llama-3.1-8b-instruct", # or alibaba/qwen3-235b-a22b
messages=[{"role": "user", "content": "Hello"}],
)Both models are live in the Antbase catalog right now — find them in the model search:
Or open them straight in the playground and run a prompt:
Providers will keep retiring endpoints, re-pricing models, and moving things behind paid tiers. Route through a pool instead of a single endpoint, and none of it is your migration.
