Providers
| Provider | Detection | Endpoint |
|---|
| OpenAI | Automatic | https://api.openai.com |
| Anthropic | Automatic | https://api.anthropic.com |
| Groq | Automatic | https://api.groq.com/openai |
| Google Gemini | Automatic | https://generativelanguage.googleapis.com/v1beta/openai |
| Azure OpenAI | X-Lectr-Provider: azure | Per-org — configured in dashboard |
https://proxy.lectr.ai/v1
OpenAI
| Model | Routing target | Recommendations |
|---|
gpt-4o | ✅ | Downgrade → gpt-4o-mini |
gpt-4o-mini | ✅ | — |
gpt-4-turbo | ✅ | Downgrade → gpt-4o |
gpt-4 | ✅ | Downgrade → gpt-4o |
gpt-4.1 | ✅ | Downgrade → gpt-4.1-mini |
gpt-4.1-mini | ✅ | Downgrade → gpt-4.1-nano |
gpt-4.1-nano | ✅ | — |
gpt-3.5-turbo | ✅ | Downgrade → gpt-4.1-nano |
o1 | ✅ | Downgrade → o1-mini |
o1-mini | ✅ | Downgrade → gpt-4o |
o3 | ✅ | Downgrade → o3-mini |
o3-mini | ✅ | Downgrade → gpt-4o-mini |
o4-mini | ✅ | Downgrade → gpt-4o-mini |
Anthropic
| Model | Routing target | Recommendations |
|---|
claude-3-opus-20240229 | ✅ | Downgrade → claude-3-5-sonnet-20241022 (conservative) or claude-3-5-haiku-20241022 (aggressive) |
claude-3-5-sonnet-20241022 | ✅ | Downgrade → claude-3-5-haiku-20241022 |
claude-3-sonnet-20240229 | ✅ | Downgrade → claude-3-5-haiku-20241022 |
claude-3-5-haiku-20241022 | ✅ | — |
claude-3-haiku-20240307 | ✅ | — |
Anthropic model names include a date suffix (e.g. 20241022). Lectr also uses
prefix matching — new Claude models are detected automatically even if not yet
in this list.
Groq
| Model | Routing target | Recommendations |
|---|
llama-3.2-90b-vision | ✅ | Downgrade → llama-3.1-70b-versatile |
llama-3.1-70b-versatile | ✅ | Downgrade → llama-3.1-8b-instant |
llama-3.1-8b-instant | ✅ | — |
mixtral-8x7b-32768 | ✅ | Downgrade → llama-3.1-8b-instant |
gemma2-9b-it | ✅ | Downgrade → llama-3.1-8b-instant |
Google Gemini
| Model | Routing target | Recommendations |
|---|
gemini-1.5-pro | ✅ | Downgrade → gemini-1.5-flash |
gemini-1.5-flash | ✅ | — |
gemini-2.0-flash | ✅ | Downgrade → gemini-1.5-flash |
Azure OpenAI
Azure uses deployment names as model identifiers — these are org-specific and
not in the detection registry. Always use X-Lectr-Provider: azure and configure
your endpoint in the dashboard.
See Multi-Provider Setup → Azure for setup instructions.
Model not listed?
Lectr uses prefix matching as a fallback — so new model versions from known
providers are usually detected correctly even before this list is updated.
If a model isn’t detected:
- Use
X-Lectr-Provider to set the provider manually
- The request goes through — detection failure never blocks traffic
- Cost tracking shows
$0.00 (pricing unavailable) until the model is added
Model list and pricing are updated regularly. If you’re missing a model you
use in production, let us know.
Routing targets
The “Routing target” column indicates whether a model can be used as the
destination in a routing rule. All listed models are valid routing targets.
Routing rules are coming in a future release.
Recommendation aggressiveness
Recommendations stay within provider — Lectr won’t suggest switching from
Anthropic to OpenAI. Within a provider, the downgrade aggressiveness depends
on task type:
| Task type | Aggressiveness |
|---|
classification | Aggressive — smallest viable model |
extraction | Aggressive (short prompts) / Moderate (long prompts) |
summarisation | Moderate (short prompts) / Conservative (long prompts) |
generation | Conservative — relies on other signals |
reasoning | Never recommended for downgrade |
