voyage-3-m-exp
Voyage-3-m-exp is an intermediate embedding snapshot trained for tasks similar to MTEB. Its training includes MTEB training splits.
Research release
Model design
The card lists 6,918 million parameters, a 32,000-token input limit and 2,048 output dimensions. It documents API access by this model name, but the main catalog does not list it. The repository publishes its tokenizer and evaluation results, not model weights.voyage-3-m-exp tokenizer, evaluation results and API model card
Practical use
Practical guidance. Use the disclosed training setup when interpreting leaderboard scores. Performance on related benchmark tasks is not independent evidence for a new application. Confirm account access before planning to use this experimental identifier.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | API |
| Published access evidence | API named in model card; absent from the main catalogvoyage-3-m-exp tokenizer, evaluation results and API model card |
| Provider guidance | Recommends voyage-3-large for production usevoyage-3-m-exp tokenizer, evaluation results and API model card |
Related models
- voyage-3-large
Voyage AI ยท Voyage 3
General embeddings with selectable dimensions and quantized output.
Launch date API