Back to Models

Mixtral 8x7B

Mistral
Released: 2023-12-11
Type: Open Source (MoE)
Context
32k
Max Out
32.0k
Cutoff
2023
Params
47B (Active 13B)

Evolution: What Changed?

  • Mixture of Experts architecture
  • 32k context window
  • Reasoning boost
Compared to mistral-7b

Evolution: What Changed?

  • Mixture of Experts architecture
  • 32k context window
  • Reasoning boost
compared to previous generation

The Breakdown

The Mixtral 8x7B represents a significant leap forward in Mistral's lineup. Released in 2023-12-11, it targets developers and enterprise use cases with a specific focus on reasoning capabilities over raw conversational speed. While previous iterations in the Open Source (MoE) category often struggled with complex multi-step instruction following, this model introduces a refined architecture that dramatically improves adherence to system prompts and reduces hallucination rates in technical domains. It competes directly with top-tier frontier models but carves out a distinct niche for workflows where precision and context retention matter more than creative flair. For businesses looking to integrate reliable AI agents, Mixtral 8x7B offers a compelling balance of performance and cost-efficiency.

Overall Score
9.1
/10
Pricing (per 1k tokens)
Input$0
Output$0
Currency: USD

The Good

  • First open MoE that worked well
  • GPT-3.5 killer
  • Fast inference

The Bad

  • Complex to host locally

The Verdict

Proved that Mixture of Experts was the future for open models.

Performance Benchmarks

mmlu
70.6
math
28.4
human eval
40.2