All tasks / Models
harrier-oss-v1-27b
Microsoft's MIT-licensed open-weights multilingual text embedding models, in 27B, 0.6B and 270M sizes.
- Microsoft
- Open weights
- Released Mar 2026
Where it ranks
- SEmbeddings & retrievalModels · Local · 74.3% on MTEB(Multilingual, v2)
Facts
- Vendor
- Microsoft
- Family
- Harrier OSS
- Released
- Mar 2026
- Weights
- Open: download and run it yourself
- License
- MIT
- Parameters
- 27.01B
- GPU memory to run locally
- ~54 GB (Derived: 27.0B parameters (Hugging Face safetensors count) x 2 bytes = ~54 GB of bf16 weights; the board lists ~51.5 GB memory use.)
- Access
- Free (open weights)
- Last verified
- 19 Oct 2026
Benchmark results
| Benchmark | Result | Entry on the leaderboard | As of | Source |
|---|---|---|---|---|
| MTEB(Multilingual, v2) | 74.3% | microsoft/harrier-oss-v1-27bMean (Task) over all 131 tasks; board rank 1 (Borda); board zero-shot 78%; read from the leaderboard's public API. | 19 Oct 2026 | mteb-leaderboard-backend.hf.space |
Where you can use it
No tracked tool lists this model yet; as open weights it runs in local runtimes.