IndexTeam released preview weights, code, an inference client and author evaluations.
IndexTeam opens a 35B MoE model for constrained translation
IndexTeam released preview weights for Index-Translate-35B-A3B, an Apache-2.0 multilingual translation model with 35B total and 3B active parameters. It supports 150 languages and instruction constraints such as terminology, style and JSON output. Code and an inference client are public, but the benchmarks are author-reported and no hosted inference provider is listed.
Availability: preview weights, code and inference client; no hosted provider
What this means for you
For localization teams, instruction-following translation can keep terminology and structured output in one pass instead of adding post-processing. Public weights permit direct testing, while the preview label and vendor-only evidence argue against production adoption.
Public artifacts and validated results are different.
The recorded fact is the public work and its artifacts. Author-reported outcomes do not establish independent reproduction. Next checks include usable code, task conditions and third-party validation.
Evidence & related material
Technical details & scores
| Architecture | Qwen3.5-derived MoE; 35B total, 3B active |
|---|---|
| Languages | 150 |
| Constraints | terminology, style, formatting and JSON output |
| License | Apache-2.0 |
| Availability | preview weights, code and inference client; no hosted provider |
| Author Results | FLORES COMET22 0.8794; instTrans instruction-following 0.8336; 2.4% off-target rate on low-resource languages |