Test of "Fastest harness"
OpenAI Agent SDK averaged 121.7s per run.
Opencode 256.2s. Claude Agent SDK 281.1s.
OpenAI was faster on all 6 runs. Not most - all.
Widest gap was Run 5: 84.7s against 494.3s.
Tested 3 agent harnesses: take a raw screen recording & polish it
Claude Agent SDK - Leans on prompt caching for long tool-use runs
OpenAI Agent SDK - Ran without caching in our setup
Opencode - very model-agnostic
@OpenAI ran in both @claudeai & @opencode
Gemini Flash TTS is ~$0.012 per 1,000 characters.
ElevenLabs is roughly 4x that.
Gemini also scores higher on the quality leaderboards.
every team building with voice runs into this math. we did too.
so we started with a blind test.
but Here's what nobody's measuring 👇
UGC is changing.
AI video is changing.
Nobody really knows what happens when they collide.
So we’re getting the people closest to it in one room. 🧃🤖
https://t.co/mpBOypBQbl
#UGC#AIVideo#CreatorEconomy
UGC IRL Mixer × AI Video
Powered by @sharewithvelo 🚀
Creators.
Marketers.
Founders.
Brand builders.
No panels.
No decks.
No corporate networking theatre.
Just: what’s working, what’s dead, what actually converts, where AI fits next!
@thatcasualvc@teerkus 🤝