Jordan Sitter rebuild — model comparison
Every build below was made from the same brief, generated from
measurements of jordansitter.com and
byte-identical across models. Each then iterated against the same automated gate
until it passed, stalled, or hit the round cap.
Cost and time are the model's own. Interventions count every retry, transport
failure, reasoning cap and truncation — a combination needing a human twice per
site is not viable at any price, so it is never folded into the cost.
Opus 5 (baseline)
- Cost
- $0.0000
- Model time
- 0s
- Gate
- not gated
- Interventions
- 0
- Outcome
- —
Homepage Job board
Gemini 3.7 Flash
- Cost
- $0.2791
- Model time
- 513s
- Gate
- home 25/5 · jobs 16/12
- Interventions
- 0
- Outcome
- CAPPED / CAPPED
Homepage Job board
Kimi K3
- Cost
- $1.4505
- Model time
- 760s
- Gate
- home 24/9 · jobs 15/13
- Interventions
- 7
- Outcome
- CAPPED / STALLED
Homepage Job board
Qwen 3.8 Max
- Cost
- $1.4874
- Model time
- 4198s
- Gate
- home 24/9 · jobs 16/12
- Interventions
- 11
- Outcome
- CAPPED / STALLED
Homepage Job board
Grok 4.6
- Cost
- $1.6313
- Model time
- 2578s
- Gate
- home 25/4 · jobs 15/10
- Interventions
- 1
- Outcome
- CAPPED / STALLED
Homepage Job board