Open Models Caught Up to Commercial Ones. They Save Money, Not Time

Cheap models have caught up with the giants

According to a Mozilla report, open and Chinese models have caught up with the so-called frontier models from major Silicon Valley players. The capability gap, for which huge sums were previously paid, has closed. Open models now offer comparable performance at a fraction of the cost.

The commercial advantage has shrunk to four months

The report shows that companies paying for the most expensive closed models today are buying roughly a four month head start before open source offers the same level at a fifth of the price. This changes the calculus for startups and the enterprise sector. Paying a premium only makes sense if you can translate those few months of head start into real business.

Metrics don't capture production reliability

Even if models match up in benchmarks, open availability doesn't solve infrastructure. Hosting, security guarantees, and API stability are reasons why large companies continue to pay OpenAI or Anthropic. Lab performance does not equal zero downtime operations.

Inference cost will decide, not charts

Price pressure will force commercial model creators to find new revenue streams, for example, by shifting to agentic systems or deeper integration into enterprise software. We should watch whether the major players start discounting their best models, or if they segment the market into cheap routine and expensive complex reasoning.

Lilith's verdict

Paying for the best model no longer means buying inaccessible magic, just a quarterly subscription to slightly better brute force.