Haiku 5.5 low leads on quality: 20.1 points higher; GPT 6 Luna low leads on cost and agent time: 4.7× cheaper and 1.4× faster.
Stet score
GPT 6 Luna lowRank 14 of 14
21.295% range 12.5–31.2
Haiku 5.5 lowRank 10 of 14
41.395% range 34.8–47.820.1 points higher
GPT 6 Luna low solved 0 tasks that Haiku 5.5 low didn’t, and Haiku 5.5 low solved 6 that GPT 6 Luna low didn’t. Both solved 5; neither solved 12.
stet-5276aa3fPassed testsstet-commit-bbbbae09Passed testsstet-66762b4cEquivalent to the merged fixstet-commit-f0caded9Equivalent to the merged fixstet-commit-5d766ddeEquivalent to the merged fixstet-43493c52Passed testsA task counts as solved when its tests passed or a judge found the fix equivalent to the merged change, as in the leaderboard’s Solved column. Stet score also weighs code quality, so it can order the two differently. 1 of the 24 tasks is left out because one of the two has no scored result on it.
| Task | GPT 6 Luna low | Haiku 5.5 low | Solved by |
|---|---|---|---|
stet-3169aefa | Passed tests | Passed tests | Both |
stet-89dfbc27 | Passed tests | Passed tests | Both |
stet-b28f4fe7 | Passed tests | Passed tests | Both |
stet-232b50ba | Passed tests | Passed tests | Both |
stet-5276aa3f | Failed | Passed tests | Haiku 5.5 low only |
stet-commit-bbbbae09 | No change made | Passed tests | Haiku 5.5 low only |
stet-15439c21 | Passed tests | Equivalent to the merged fix | Both |
stet-66762b4c | Failed | Equivalent to the merged fix | Haiku 5.5 low only |
stet-commit-f0caded9 | Failed | Equivalent to the merged fix | Haiku 5.5 low only |
stet-commit-5d766dde | No change made | Equivalent to the merged fix | Haiku 5.5 low only |
stet-43493c52 | Failed | Passed tests | Haiku 5.5 low only |
stet-6f84e978 | Failed | Failed | Neither |
stet-commit-f8ae1177 | No change made | Failed | Neither |
stet-2450ca2d | Failed | Failed | Neither |
stet-commit-342d0ac0 | Failed | Failed | Neither |
stet-commit-5e075641 | Failed | Failed | Neither |
stet-commit-2826765e | Failed | Failed | Neither |
stet-commit-a8716b88 | Failed | Failed | Neither |
stet-commit-30506f8e | Failed | Failed | Neither |
stet-commit-87814583 | Failed | Failed | Neither |
stet-commit-d2ab0c25 | Failed | Failed | Neither |
stet-commit-e41c2e2d | Failed | Failed | Neither |
stet-commit-e928166f | Failed | Failed | Neither |
| Metric | GPT 6 Luna low | Haiku 5.5 low |
|---|---|---|
| Rank | 14 of 14 | 10 of 14 |
| Stet score | 21.295% range 12.5–31.2 | 41.395% range 34.8–47.8 |
| Solved | 5/245 passed, 0 judged equivalent | 11/247 passed, 4 judged equivalent |
| Passed review | 5/204 not recorded | 8/24 |
| Matched merged change | 5/204 not recorded | 9/24 |
| Cost / task | $0.05 | $0.25 |
| How cost was recorded | reported by its harness | reported by its harness |
| Agent time / task | 6.9 min | 9.9 min |
| Ran in | Daytona sandbox | Daytona sandbox |
The leaderboard’s readings on groups of tasks. Each group holds fewer tasks than the overall standings, so treat these as leads to test on your repo.
11 tasks
11 tasks whose merged change was 2–149 lines in 1–2 files. The 7 small fixes below are among them.
Haiku 5.5 low matched the merged change more often: 7 of 11 to GPT 6 Luna low’s 5.
| GPT 6 Luna low | Haiku 5.5 low | |
|---|---|---|
| Matched merged change | 5/11 | 7/11 |
| Passed review | 5/11 | 7/11 |
| Cost / task | $0.02 | $0.02 |
| Agent time / task | 2.1 min | 2.8 min |
7 tasks
7 tasks whose instructions start with “Fix”. Their merged changes were 4–29 lines in 1–2 files.
Both matched the merged change on 5 of 7.
| GPT 6 Luna low | Haiku 5.5 low | |
|---|---|---|
| Matched merged change | 5/7 | 5/7 |
| Passed review | 5/7 | 5/7 |
| Cost / task | $0.01 | $0.01 |
| Agent time / task | 2.1 min | 2.7 min |
13 tasks
13 tasks whose merged change ran over 150 lines or more than 3 files: 113–1,420 lines in 2–19 files.
| GPT 6 Luna low | Haiku 5.5 low | |
|---|---|---|
| Matched merged change | 0/94 not recorded | 2/13 |
| Passed review | 0/94 not recorded | 1/13 |
| Cost / task | $0.08 | $0.45 |
| Agent time / task | 11.0 min | 15.9 min |
Stet replays your merged PRs, scores quality above pass/fail, and delivers comparison reports through your agent.
Run it on your repocompare-n24-r3b.