Get Started

Meta researchers taught an 8B AI model to match Claude Opus 4.5 — without the frontier price tag

ResearchAI Agents

First reported · 57d ago · 1 source

Crest summary — written by AI from the 1 source below; every figure is checked against them · model deepseek-chat

Summary

Meta researchers trained an 8B AI model to match Claude Opus 4.5 on complex enterprise agent workflows, according to VentureBeat. The work focuses on the runtime harness rather than the model's context window.

Key facts

  • Meta researchers trained an 8B AI model
  • The model matched Claude Opus 4.5, per VentureBeat
  • Reported on 2026-08-28 by VentureBeat
  • Work targets long-running enterprise agent workflows
  • Approach relies on the runtime harness layer

Why it matters

Teams building agents may get comparable results from a much smaller model, which could lower the cost of running long enterprise workflows. The harness, not just the model, becomes the place to invest engineering effort.

Companies

Technologies

  • Claude Opus 4.5model

Original sources (1)

The report this event was built from. Crest of Tech keeps no article bodies: the full text is at each source.

  1. VentureBeatPublication · Primary
    Meta researchers taught an 8B AI model to match Claude Opus 4.5 — without the frontier price tag · By bendee983@gmail.com (Ben Dickson)
    Read at VentureBeat
Meta researchers taught an 8B AI model to match Claude Opus 4.5 — without the frontier price tag — Crest of Tech