Claude Fable 5 vs GPT-5.6 Sol: $100 AI-Driven Music Video Showdown

Imagine handing an AI model a song, a budget, and a set of production tools, then watching it deliver a complete music video on its own. That’s exactly what we did with Claude Fable 5 and OpenAI’s GPT‑5.6 Sol, each given the hit “Uptown Funk” and a $100 spend limit. The models ran an autonomous loop of planning, web search, budget checks, image/video generation, and ffmpeg editing. Both completed the task without human intervention, producing a full‑length video with the original track. This experiment shows how far autonomous tool‑calling has come and what cost trade‑offs creators face.

What happened

We built a lightweight agentic harness that hands a model a song, a hard dollar budget, and six tools: a planning stub, web_search, get_budget, generate_image, generate_video, and a local run_command with ffmpeg. Each model received the same lyric‑timestamped description of Bruno Mars and Mark Ronson’s “Uptown Funk” and was allowed to decide how to research video models, generate clips, edit, and assemble the final cut. We ran four experiments: Claude Fable 5 and GPT‑5.6 Sol at $25 and $100 budgets. All runs finished on their own and produced a valid video file. At the $100 level the models spent the allocated generation budget—$48.60 for Claude Fable 5 and $36.57 for GPT‑5.6 Sol—demonstrating that more money translates into more generated footage, though the spend does not include the compute cost of the language model itself. The full logs, tool‑call transcripts, and the open‑source harness are available at github.com/hershalb/music-video‑arena, allowing anyone to replicate or extend the experiment.

Why it matters

For creators, the ability to hand off an entire video production to an LLM reduces manual editing time and opens new rapid‑prototyping workflows. However, the cost differential—Claude Fable 5’s higher token price and higher generation spend—means that budget‑constrained teams may favor GPT‑5.6 Sol despite slightly shorter footage. The experiment also highlights the importance of tool selection; models that can reliably invoke external generation services and manage ffmpeg commands autonomously are better suited for end‑to‑end media pipelines.

How to think about it

Start by defining the creative output and the maximum spend you’re willing to allocate for generation services. Choose a model whose API pricing aligns with that budget—GPT‑5.6 Sol currently offers about a 50 % cost advantage over Claude Fable 5. Next, map the required toolchain (search, generation, editing) and verify that the model can invoke each tool reliably. Run a short pilot with a $25 budget to gauge footage length and quality before scaling to $100 or more. Finally, factor in hidden compute costs for the language model itself; they can double the effective spend if not accounted for.

FAQ