E5-Mistral 7B Instruct
Instruction-tuned 7B embedding model that first put LLM decoders atop the MTEB leaderboard
About
E5-Mistral 7B Instruct adapts the Mistral 7B decoder into a text embedding model, using task-specific instruction prefixes to steer the representation for retrieval, clustering, and classification. Released in January 2024, it was the first decoder-only LLM embedding model to top the MTEB leaderboard, kicking off the wave of 7B-class embedding models. It produces 4096-dimensional vectors and works well for high-quality English retrieval when you can afford the GPU. The tradeoff is cost: at 7 billion parameters it needs serious VRAM and encodes noticeably slower than BERT-class models, so it suits quality-first pipelines rather than high-throughput or CPU-only setups.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
MTEB average ~66.6 (reported in third-party comparison tables)
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (verified tier, metrics-v1). Evidence: Architecture, 4096-dim vectors, MIT license, and Jan 2024 first-to-top-MTEB claim corroborated across three independent comparison tables read this session.; MTEB ~66.6 figure comes from third-party tables, not the original paper — treated as approximate.; Context window 4,096 tokens from comparison tables; the base Mistral 7B supports longer windows, so this may understate the model's capability..