Map/Models and thresholds/Wave 6

Firm-owned post-trained models

Firms post-train and keep their own models on their own data instead of renting a general frontier model, often while still renting the compute.

capabilityEmerging Open on the map →

Facts

Lane: Models and thresholds
Wave: 6 of 12 (build order, not a date)
Needs: 12 boxes all the way back through hard links
Unlocks: 0 boxes downstream through hard links
Leverage: None
Cluster: Agents at work
Found by: ownership inversion lens

Evidence as of October 2026

By Thinking Machines' account (15 Oct 2026), Bridgewater's text-to-SQL model built on Tinker beat named frontier models at 15% of their cost; YC's Winter 2027 'Own Your Intelligence' request backs the same shift.

Around it

Needs · 3

Open-weight modelscapability · Built today
RL on checkable taskscapability · Built today
Company data connectorscapability · Built today

Unlocks · 3

Knowledge-rich agentsstage · Emerging · helps, not required
Continual-learning agentsstage · Emerging · helps, not required
AI-native application revenuecapability · Built today · helps, not required

Loops and off-ramps

Learning from deploymentreinforcing loop
Capex pullbackoff-ramp · cut off
Taiwan chip disruptionoff-ramp · cut off
Power shortfalloff-ramp · cut off
Capability plateauoff-ramp · cut off
Training data ruled outoff-ramp · slowed
Misuse catastropheoff-ramp · cut off

Openings researchers' judgement, October 2026 scouting

Startup room 52/100 · rank 32 of 124 Research room 39/100 · rank 94 of 124

Crowding4 Lab overlap3 Small-team fit3.5 Buyer today3.5 Research openness3.5

Depth: light
Players: Thinking Machines Tinker; River AI ($1.1B, Aug 2026); Prime Intellect ($130M); Deep Cogito ($43M A); Bespoke Labs ($40M); Together post-training; OpenAI RFT
Wedge: Sell the step platforms skip: turn one vertical's workflow logs (e.g. asset-manager SQL and document tasks) into graders and evals, then post-train on rented Tinker-style platforms. startup