
Harness benchmark / 2026-11-04
GPT-5.6 Terra Custom skills supplied: None
Time taken
Execution time, seconds
Cost
Priced token subtotal, USD · * incomplete · — unknown
| Harness | Run | Total | Input tokens | Output tokens | Cache read | Reasoning | Original preview | Publication check | ||
|---|---|---|---|---|---|---|---|---|---|---|
| Codex | Complete | 49.9s | 59.4s | 16,820 | 5,415 | 53,737 | $0.1094 | Medium | Published | Add + persistence passed |
| OpenCode | Complete | 59.5s | 188.6s | 29,578 | 5,949 | 224,185 | $0.1734 | Not exposed | Publishing failed | Add + persistence passed |
| Hermes | Complete | 80.8s | 234.1s | 25,053 | 5,735 | 228,921 | $0.1647* | Not exposed | Published | Startup error |
| Prime Agent | Complete | 84.9s | 280.7s | 35,656 | 6,788 | 287,166 | $0.2102 | Medium | Published | Add + persistence passed |
| Pi | Complete | 94.9s | 105.2s | 57,594 | 8,294 | 460,836 | $0.3069* | Medium | Published | Add + persistence passed |
| JetBrains Junie | Complete | 100.4s | 124.8s | 49,026 | 10,532 | 296,871 | $0.2838 | Not exposed | Publishing failed | Add + persistence passed |
| Cursor Cloud | Complete | 153.4s | 216.0s | Unknown | Unknown | Unknown | Unknown | Not exposed | No public URL | Add + persistence passed |
| Factory Droid | Complete | 168.4s | 184.9s | 37,097 | 26,000 | 439,953 | $0.4742* | Not exposed | Publishing failed | Add + persistence passed |
Sorted by time taken, fastest first.


OpenCode
59.5s · $0.1734
Add + persistence passed
Agent reported 2 passing tests. Usage includes GPT-5.4 Nano: 1,037 input and 13 output tokens. Public preview needed separate recovery.

Hermes
80.8s · $0.1647*
Startup error
Agent reported static core-flow and HTTP checks. Publication check: fresh storage triggers “Cannot access save before initialization”. Original app preserved.

Prime Agent
84.9s · $0.2102
Add + persistence passed
Agent reported syntax and local checks. Usage was provisional in the first summary and has since settled.

Pi
94.9s · $0.3069*
Add + persistence passed
Agent reported 3 passing tests. Editing uses native prompt dialogs.

JetBrains Junie
100.4s · $0.2838
Add + persistence passed
Agent reported 4 passing tests. Public preview needed separate recovery.

Cursor Cloud
153.4s · Unknown
Add + persistence passed
Agent reported syntax and local HTTP checks. Used Cursor’s native cloud environment. Usage returned zeros and is treated as unknown. Public preview needed separate recovery.

Factory Droid
168.4s · $0.4742*
Add + persistence passed
Agent reported syntax and local HTTP checks. Public preview needed separate recovery.