Layer 01 · Shared model

One model,for the whole team.

Not fifteen personal accounts nobody can see, and it is the one layer we never sell you. The licence is in your name.

See the shared stackBook a discovery call
The layerWhere it runs

Three places,one axis.

How much of your work leaves your perimeter is the only question this layer decides — and you can change your mind later without rebuilding anything above it.

Three places the model can runThree deployment positions on one axis, ordered by how much leaves your perimeter: frontier endpoints on the provider’s infrastructure, open-weight frontier models served in a region you choose, and fully on-premises on your own hardware. Moving between them is a configuration change rather than a rebuild.MOST CAPABLE, LEAST CONTAINEDMOST CONTAINED01Frontier endpointsThe best models there are,under a signed agreementTHEIR INFRASTRUCTURE02Open weightsFrontier-class open models,served in a region you nameWHERE YOU SAY03On-premisesWhen nothing may leavethe building at allYOUR OWN HARDWAREMOVE LATER — IT IS A CONFIG CHANGELEGENDWhere most Canadian teams landThe licence is in your name at every stop
What we have published

We benchmark thisin the open.

The facts on this layer move monthly, so we write them down with the numbers attached rather than summarising them here.

11 min readRunning Open Models at Frontier LevelOpen weights now land within a few points of the best proprietary models on published benchmarks. Closing the last stretch is a systems problem, and it is one you can solve.13 min readThe Best Open-Source Coding Models in 2026 (and How They Compare to Claude)A practical look at the best open-weight coding models in 2026: Kimi K3, GLM-5.3, GLM-5.3-Flash, DeepSeek-V4-Pro, GLM-5.2, and how they compare to Claude Sonnet and Opus.9 min readThe Cheapest Model Is the One That Finishes: Cost Per Solved TaskPrice per million tokens tells you what an attempt costs. Divide by how often the model succeeds and you get the number that shows up on the invoice.6 min readWhere Does Your Code Go? A Guide to AI Inference Data ResidencyAI inference data residency explained: what it means, why it matters for your prompts and source code, and how to control where inference actually runs.7 min read88.2 and 28.3 Are the Same Model: Read the Version NumberOne model card reports 88.2 on Terminal-Bench 2.1 and 28.3 on Terminal-Bench 3.0. Same model, same day. Here is how to read benchmark numbers so they mean something.
Bring your constraints

Book a discovery call.

Start with the map of your organization, or with the one job that hurts. Measured in hours and money, and everything we build stays yours.

Loading form…