Skip to the comparison
STET
Get a free trial

GPT 6 Luna low vs Haiku 5.5 low+guidance

Haiku 5.5 low+guidance leads on quality: 21.2 points higher; GPT 6 Luna low leads on cost and agent time: 5.4× cheaper and 1.8× faster.

24 tasksData from

Stet score

GPT 6 Luna lowcodex · low reasoningRank 15 of 15

21.295% range 12.5–31.2

Haiku 5.5 low+guidanceclaude-code · low reasoningRank 10 of 15

42.495% range 33.2–51.721.2 points higher

Their 95% score ranges don’t overlap.

Cost / task

GPT 6 Luna low$0.055.4× cheaper
Haiku 5.5 low+guidance$0.29

Agent time / task

GPT 6 Luna low6.9 min1.8× faster
Haiku 5.5 low+guidance12.4 min

Quality

Head-to-head matchups

GPT 6 Luna low won3

Tied51 undecided

Haiku 5.5 low+guidance won15

Haiku 5.5 low+guidance’s win share 80%, 95% range 65–92%Ties are within judge noise.

Task by task

GPT 6 Luna low solved 1 task that Haiku 5.5 low+guidance didn’t, and Haiku 5.5 low+guidance solved 5 that GPT 6 Luna low didn’t. Both solved 4; neither solved 13.

GPT 6 Luna low only 1

  • stet-15439c21Passed tests

Haiku 5.5 low+guidance only 5

  • stet-5276aa3fPassed tests
  • stet-commit-bbbbae09Passed tests
  • stet-66762b4cPassed tests
  • stet-commit-f0caded9Equivalent to the merged fix
  • stet-43493c52Passed tests

A 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.

All 23 compared tasks
Task by task
TaskGPT 6 Luna lowHaiku 5.5 low+guidanceSolved by
stet-3169aefaPassed testsPassed testsBoth
stet-89dfbc27Passed testsPassed testsBoth
stet-b28f4fe7Passed testsPassed testsBoth
stet-232b50baPassed testsPassed testsBoth
stet-5276aa3fFailedPassed testsHaiku 5.5 low+guidance only
stet-commit-bbbbae09No change madePassed testsHaiku 5.5 low+guidance only
stet-66762b4cFailedPassed testsHaiku 5.5 low+guidance only
stet-15439c21Passed testsFailedGPT 6 Luna low only
stet-commit-f0caded9FailedEquivalent to the merged fixHaiku 5.5 low+guidance only
stet-commit-5d766ddeNo change madeFailedNeither
stet-43493c52FailedPassed testsHaiku 5.5 low+guidance only
stet-6f84e978FailedFailedNeither
stet-commit-f8ae1177No change madeFailedNeither
stet-2450ca2dFailedFailedNeither
stet-commit-342d0ac0FailedFailedNeither
stet-commit-5e075641FailedFailedNeither
stet-commit-2826765eFailedFailedNeither
stet-commit-a8716b88FailedFailedNeither
stet-commit-30506f8eFailedFailedNeither
stet-commit-87814583FailedFailedNeither
stet-commit-d2ab0c25FailedFailedNeither
stet-commit-e41c2e2dFailedFailedNeither
stet-commit-e928166fFailedFailedNeither

Side by side

Side by side
MetricGPT 6 Luna lowHaiku 5.5 low+guidance
Harnesscodexclaude-code
Reasoning effortlowlow
Rank15 of 1510 of 15
Stet score21.295% range 12.5–31.242.495% range 33.2–51.7
Solved5/245 passed, 0 judged equivalent9/248 passed, 1 judged equivalent
Cost / task$0.05$0.29
Agent time / task6.9 min12.4 min

See which setup wins on your repo

Stet replays your merged PRs, scores quality above pass/fail, and delivers comparison reports through your agent.

Run it on your repo

About this data

  • 24 tasks, each a merged change replayed from the repo as it was just before the merge.
  • Data from .
  • Both setups come from the same leaderboard run of 15 setups: bmk_d4ba310fdf50db4ccb2092e3.

See all 15 setups on the leaderboardHow Stet measures