GPT 5.6 Sol Made This Entire Video

Video thumbnail: GPT 5.6 Sol Made This Entire Video
Jul 9, 20265m 24s video lengthNate Herk | AI Automation

The Signal

A creator demonstrated that GPT 5.6 Soul on the 'Ultra' setting can autonomously coordinate a full video production—including research, scripting, rendering, and QC—from a single prompt. While OpenAI markets this as their strongest coding model, the workflow highlights a tension between high-effort, agentic delegation and the substantial token overhead that follows.

The Case

  • Nate, the creator's partner, did not record, edit, or review the video, illustrating the model's capacity to bypass human production steps by orchestrating external tools like 11 Labs, Hen, and Hyperframes.0:07
  • To manage motion rendering, the model used browser automation to force a specific engine setting ('Avatar V') because the API could not reliably lock it, underscoring that 'end-to-end' production currently requires heavy infrastructure layering.1:50
  • The speaker estimated the production cost at approximately $300, driven by a total of 450 million tokens across the main agent and nine spawned sub-agents, though they suspect the 'Ultra' setting overdelegated tasks unnecessarily.3:33
  • Despite this specific run's cost, the speaker claims GPT 5.6 Soul is generally half the price of the prior 'Fable 5' model, though this assertion remains self-reported and lacks broad, independent validation.4:11
  • The model successfully passed a 13-task internal test with a 97% objective score, and its performance on benchmarks like Terminal Bench 2.1 (91.9%) indicates genuine strength in structured execution rather than mere generative fluff.0:51

The 1 Minute Signal Take

This demo proves that models can now act as effective production orchestrators, provided you accept a 'black box' layer of automated verification and tolerate high, sometimes inefficient, token consumption. It serves as a necessary reminder that benchmark scores in isolation hide the messy, multi-tool realities of actually getting a finished product out of a model.

Pro Analysis

Why It Matters

This demo serves as a proof of concept for the migration away from 'assistant' models toward 'operator' models. By automa...

Full analysis always available on Pro.

Time saved:3m 50s

Share this

Tags

Written by: 1 Minute Signal Editorial Team