Which LLM does Creopus use?
Creopus runs a provider-fallback chain. Every tool uses the same default order:
Gemini 2.5 Flash → Cerebras → Claude Haiku 4.5
| Provider | Model | Context | Max output | Images |
|---|---|---|---|---|
Google Vertex AI (asia-south1) | Gemini 2.5 Flash | 1M input | 65K | Yes |
| Cerebras | gpt-oss-120b | 128K shared | 40K | No |
| Anthropic | Claude Haiku 4.5 | 200K shared | 64K | Yes |
If the primary provider is rate-limited, errors out, or the prompt exceeds its context window, the platform automatically retries on the next provider in the chain. Tools that require image understanding skip providers that cannot accept images.
The order is configurable per tool by environment variable, so a tool's chain can be re-ordered without a release. Each tool always keeps three providers: a shorter list is completed from the default order.