Mixtral 8x7B
The MoE breakthrough: GPT-3.5-class, Apache-2.0, 6x faster.
About
Mixtral-8x7B-Instruct-v0.1 (December 2023) was the mixture-of-experts breakthrough: 8 experts of 7B (46.7B total, 13B active per token via top-2 routing) matching GPT-3.5-class quality at 6x the inference speed — all Apache-2.0 licensed. 212.8k HF downloads, 4767 likes, not gated. It proved sparse MoE could be open and practical, spawning the whole open-MoE wave. 93GB of weights still needs serious GPUs, and the 8x22B plus newer generations supersede it — but as the MoE pioneer with a permissive license, it's a landmark.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-reported (Mistral, Dec 2023): GPT-3.5-competitive on MMLU, HumanEval, MATH at 6x inference speed — provider-reported. Historical.
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: HF downloads 212,829 (peer-normalized); released 1038 days ago. Manual review: the MoE breakthrough (8x7B, GPT-3.5-class at 6x speed), 213k dl; superseded by 8x22B+; 93GB weights need real GPUs. License: Apache-2.0.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 7.9/10
- Maintenance · recency, releases, issue load
- 4.0/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 9.0/10
- · HF downloads 212,829 (peer-normalized)
- · released 1038 days ago
- · manual review: the MoE breakthrough (8x7B, GPT-3.5-class at 6x speed), 213k dl; superseded by 8x22B+; 93GB weights need real GPUs