Three software companies shipped proprietary AI models in one week. Intercom's Apex 1.0 beats GPT-4.5 and Claude Opus 4.5 on customer support benchmarks. Chroma's Context-1 hits 97% on agent search benchmarks and is open-source under Apache 2.0. Cursor shipped one too. This is not a trend. It is a structural shift in how software competes.
The performance numbers are real but the more interesting split is strategic. Intercom CEO Eoghan McCabe is betting that proprietary models become the only durable differentiator once features are cheap to build. One gaming customer saw resolution rates climb from 68% to 75% on Apex 1.0. Chroma is playing a different game entirely: Context-1 is free to use, and the model is distribution infrastructure for their vector database business. Same week, two opposite philosophies.
Rich Sutton's Bitter Lesson warns that specialized advantages erode as general-purpose models scale. That pattern may hold. But rising GPU costs and inference prices could carve out real space for efficient, task-specific models. The piece is worth reading in full for how it maps proprietary versus open-source models as two distinct levers, one for closing deals, one for building brand, and what that means for every SaaS vendor still relying on OpenAI's API.
[READ ORIGINAL →]