Mixtral 8x22B Beats GPT-4 on Benchmarks. The License Matters More Than the Score.
Mistral has released Mixtral 8x22B Instruct under the Apache 2.0 license. The open model reportedly beats GPT-4 on several benchmarks and can be run locally or hosted for pennies per query, letting small teams avoid vendor lock-in. Well, actually, the real breakthrough here is the economics, not the benchmark numbers.
This demonstrates the principle of commoditization through open weights. When a capable model is released under Apache 2.0, it removes the toll booth that proprietary vendors depend on. The mechanism is straightforward: permissive licensing plus local execution equals zero marginal cost per query. Small teams win because they control the infrastructure.
Mistral, the French AI lab, released Mixtral 8x22B Instruct with an Apache 2.0 license. The model beats GPT-4 on several benchmarks and can be hosted for pennies per query.
- Visit the Mistral website at mistral.ai to read the Mixtral 8x22B release notes and understand the model specifications.
- Download a lightweight model runner like LM Studio or Ollama, which are free consumer tools for running open models locally on a standard laptop.
- Search for Mixtral or a smaller Mistral variant within the runner and load it, then test it with a few prompts to experience what open-weight inference feels like with no API costs.