Claude Fable 5 vs GPT-5.6 Sol: a $100 music video bake-off

The music-video-arena crew handed two models the same song (Uptown Funk), a budget cap, and six tools, including web_search, generate_video, and a shell with ffmpeg. Each model hunted down its own video generators, shot the clips, and cut the final edit on its own. Four runs total: Fable 5 and Sol at $25 and $100 budgets.
All four finished unassisted and produced a complete video. At the $100 cap, Sol spent $36.57 on generation while Fable 5 went up to $48.60 and rendered at 1920x1080. Fable’s token costs ran $16.99 to $25.05 per run; Sol stayed around $3 to $4.
Fable 5’s full $100 run came to $73.65, the priciest of the bunch. Sol’s $100 run mixed three video models in a single pass, while Fable stuck to Seedance 1.0 Pro. The harness is open source on GitHub.
Up next
Claude hit with elevated error rates across several models again
Anthropic reported elevated error rates across several models at 18:36 UTC on July 16 and opened an investigation. The incident hit claude.ai, the API, Claude Code, and Claude Cowork, following similar outages on June 16 and July 9.
Related stories
- Kimi K3 vs Fable 5: same code, a third of the price, four times slower
- Intelligence Index v5: Opus 5.5 at 58, GPT-6 Astra at 57
- Andon Labs puts GPT-6 Astra ahead of Claude Fable 5.1
- GPT-6 Astra edges past Claude Fable 5.1 on WebDev Arena
- METR clocks Anthropic benchmark gains at 1.9x old pace
- Cat Wu, Claude Code’s product lead, joins OpenAI
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.

