I think it's a MoE probably with 8b activated parameters - so not exactly 8b but maybe 8b x 8 But on many benchmarks it does surpass GPT 4o mini - some benchma

I think it's a MoE probably with 8b activated parameters - so not exactly 8b but maybe 8b x 8 But on many benchmarks it does surpass GPT 4o mini - some benchmarks are even better than GPT 4o. But yes in general it's because its a powerful open source models!