Trajectory launches post-training platform to let anyone train, eval and deploy trillion-parameter models
yenkel · x · 2026-11-03
Trajectory Labs launched a post-training platform positioned as "the foundation for continual learning." The team claims that with just a few agent prompts, users get a full suite to train, eval, and deploy trillion-parameter models.
- Two research previews ship alongside: Continual Grading — scales your taste so the model learns what good looks like from far more examples than you could grade by hand; and GS1, a new algorithm (details not yet disclosed).
- The vision: models shouldn't be trained once but keep learning on their own from the world around them.
- Yenkel (ex-OpenAI) endorsed the launch, saying he strongly believes most companies will post-train multiple models for their own use cases.
More from coding & agent
- Obsidian dev open-sources Pele, a Mermaid-compatible renderer 100-500x faster than mermaid.js — Teknium · 2026-11-03
- Agent-built eval harness spawns subagents via codemode in one durable session — mitsuhiko · 2026-11-03
- Impeccable crosses 100k GitHub stars, with weeks of releases teased ahead — jamdac · 2026-11-03
- Every to demo 'Compound Sandwich': which engineering workflows to hand to agents — every · 2026-11-03
- Every launches an AI coworker in Slack that turns tagged tasks into reusable team skills — every · 2026-11-03
- Founder runs 2 AI companies with 8 Grok bots: Chief, PA, email, ads and GTM agents — alexgoughcooper · 2026-11-03