The Watch

Two frontiers. Two watches.

There are two AI frontiers now, and they move differently — so the Watch tracks them separately. The Commercial Watch follows the proprietary labs and their API models. The Open Source Watch follows open-weight models: their labs, their licenses, their standings, and their releases.

The Commercial Watch monitoring wall
Commercial Watch

The proprietary frontier

The interactive Singularity Watch over the closed labs: a provider radar, benchmark curves plotted from cited public sources, and an incident feed. Current top of the coding curve: Claude Opus 5 at 96% and GPT-5.6 Sol at 96.2% on SWE-bench Verified — a saturated benchmark, and the curves say so.

  • Providers: OpenAI, Anthropic, Google, xAI, Meta's API models, and peers
  • Every feed labeled live or simulated in place — nothing dressed up
  • Every benchmark point carries its public source
Open the Commercial Watch
A global open-model ecosystem
Open Source Watch

The open-weight frontier

A dedicated tracker for downloadable models: the labs shipping them, the licenses that decide what you can actually ship, standings from public leaderboards, and a dated release feed. Current Index leader: Xiaomi's MiMo-V2.6-Pro; top open coding score: DeepSeek V4-Pro at 80.6% on SWE-bench Verified.

  • Labs: Xiaomi, Z.ai, Moonshot, DeepSeek, Alibaba, Mistral, NVIDIA, Meta, Google
  • Licenses recorded per model — MIT and Apache 2.0 vs. bespoke and revenue-gated
  • Standings are dated snapshots with sources, never silently mixed
Open the Open Source Watch