I work at a mid-sized startup dealing with latency issues in customer-facing flows that use LLMs. Using OSS-120B seems preferable to 5-mini or Anthropic models in many cases when we need speed, intelligence, and cost control. Is there some catch here beyond needing to acquire higher rate limits?