Ministral 8B 2410
Ministral 8B 2410 is the larger model in the original Ministraux pair. Its public instruction weights use the Mistral Research License, with separate terms needed for commercial use.
Deprecated API; check access details
Attention and vocabulary
The dense Transformer has 36 layers and alternates a 128K attention span with 32K sliding windows. It uses the V3-Tekken tokenizer with 131,072 vocabulary entries. The model card describes multilingual and code training and support for function calls.Ministral 8B 2410 official model card
Research weights are not the later open release
This checkpoint is Ministral-8B-Instruct-2410. It predates Ministral 3 8B and has different licensing. Its hosted API is deprecated, but the research download remains listed. Model names that omit the version can conceal this difference.Ministral 8B 2410 official model cardMinistral 8B 2410 model and lifecycle record
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Deprecated API; check access details |
| Launch date | October 2024 |
| Access | API, Downloadable |
| Model identifiers | Ministral-8B-Instruct-2410ministral-8b-2410 |
| Context window | 128,000 tokens The model catalog advertises 128k context for this version. This is not a measured recall limit. Ministral 8B 2410 model and lifecycle record |
| Weight license | Mistral Research License; commercial use needs separate permission/licenseMinistral 8B 2410 official model card |
| Access and lifecycle | Hosted API deprecated on 2 December 2025. Downloadable weights remain listed.Ministral 8B 2410 model and lifecycle recordMinistral 8B 2410 official model card |
| Parameters | 8,019,808,256Ministral 8B 2410 official model card |