Comment by real0mar

The inconvenient truth might be that the other models score higher than OLMO because they aren't restricted to purely "open and accessible" training data. Who knows what private or ethically dubious data went into training Mistral or llama, for example.