Where do these coding results come from?

Both coding agents worked the same 24 merged changes from the private Stet-AI/stet repository, each in the real repo as it was before the change. The repo’s tests and judges graded every patch against the merged change; no public benchmark scores are used.

Are reasoning settings combined into one model score?

No. Each effort pairing keeps its own recorded quality call, head-to-head record, cost and agent time. Incompatible scores are never averaged.

GPT 5.6 Luna vs Haiku 5.5

GPT 5.6 Luna has a slight quality edge in 1 of 4 pairings; none is a clear lead. 4 pairings; 24 merged changes in Stet-AI/stet. Compare the individual calls below to choose which setup to investigate.

What this comparison is

1 real repositoryStet-AI/stet, private; not public benchmark scores
24 merged changeseach run end to end by the coding agent from the repo as it was before the change
Graded against the merged changeby the repo’s own tests and by judges comparing each patch with it
Quality, cost and timea smaller sample than public benchmarks, but real agent work on a real codebase

Compare coding models Low vs high vs xhigh/max reasoning for coding agents

Recorded setup pairings

Each row is a separate recorded comparison. Scores are not pooled or averaged across effort settings. W/D/L means wins, draws and losses; undecided judgments remain undecided.

Stet score: A setup’s average chance of beating another setup on a task, from 0 to 100; 50 is even with the field.

Setup pairing	Recorded quality call and record	Quality evidence	Cost per task	Agent time per task
GPT 5.6 Luna max vs Haiku 5.5 maxGPT 5.6 Luna max · max reasoningHaiku 5.5 max · max reasoning23/24 tasks observed by both setups.	No clear quality difference: GPT 5.6 Luna max won 2, Haiku 5.5 max won 2, 18 tied, 2 undecided; Haiku 5.5 max leads on cost: 1.2× cheaper; GPT 5.6 Luna max leads on agent time: 2.0× faster.Head to head, GPT 5.6 Luna max won 2, tied 18 and lost 2 against Haiku 5.5 max, with 2 undecided.	No clear difference 2–2 head to head, 18 tiedGPT 5.6 Luna max: 49.7 (95% range 44.0–55.6)Haiku 5.5 max: 48.2 (95% range 38.8–58.1)	GPT 5.6 Luna max: $0.90 Haiku 5.5 max: $0.78	GPT 5.6 Luna max: 28.5 min Haiku 5.5 max: 58.4 min
GPT 5.6 Luna max vs Haiku 5.5 lowGPT 5.6 Luna max · max reasoningHaiku 5.5 low · low reasoning24/24 tasks observed by both setups.	GPT 5.6 Luna max has a slight quality edge, 2–0 head to head with 21 tied and 1 undecided; Haiku 5.5 low leads on cost and agent time: 14.0× cheaper and 2.9× faster.Head to head, GPT 5.6 Luna max won 2, tied 21 and lost 0 against Haiku 5.5 low, with 1 undecided.	Slight edge 2–0 head to head, 21 tiedGPT 5.6 Luna max: 49.7 (95% range 44.0–55.6)Haiku 5.5 low: 44.7 (95% range 39.5–49.1)	GPT 5.6 Luna max: $0.90 Haiku 5.5 low: $0.06	GPT 5.6 Luna max: 28.5 min Haiku 5.5 low: 9.9 min
GPT 5.6 Luna low vs Haiku 5.5 maxGPT 5.6 Luna low · codex · low reasoningHaiku 5.5 max · max reasoning23/24 tasks observed by both setups.	No clear quality difference: Haiku 5.5 max won 3, GPT 5.6 Luna low won 2, 17 tied, 2 undecided; GPT 5.6 Luna low leads on cost and agent time: 15.3× cheaper and 8.7× faster.Head to head, Haiku 5.5 max won 3, tied 17 and lost 2 against GPT 5.6 Luna low, with 2 undecided.	No clear difference 2–3 head to head, 17 tiedGPT 5.6 Luna low: 44.6 (95% range 38.3–50.3)Haiku 5.5 max: 48.2 (95% range 38.8–58.1)	GPT 5.6 Luna low: $0.05 Haiku 5.5 max: $0.78	GPT 5.6 Luna low: 6.7 min Haiku 5.5 max: 58.4 min
GPT 5.6 Luna low vs Haiku 5.5 lowGPT 5.6 Luna low · codex · low reasoningHaiku 5.5 low · low reasoning24/24 tasks observed by both setups.	No clear quality difference: GPT 5.6 Luna low won 1, Haiku 5.5 low won 1, 21 tied, 1 undecided; GPT 5.6 Luna low leads on cost and agent time: 1.3× cheaper and 1.5× faster.Head to head, Haiku 5.5 low won 1, tied 21 and lost 1 against GPT 5.6 Luna low, with 1 undecided.	No clear difference 1–1 head to head, 21 tiedGPT 5.6 Luna low: 44.6 (95% range 38.3–50.3)Haiku 5.5 low: 44.7 (95% range 39.5–49.1)	GPT 5.6 Luna low: $0.05 Haiku 5.5 low: $0.06	GPT 5.6 Luna low: 6.7 min Haiku 5.5 low: 9.9 min

Questions about these results

Which is better for coding, GPT 5.6 Luna or Haiku 5.5?: GPT 5.6 Luna has a slight quality edge in 1 of 4 pairings; none is a clear lead. 4 pairings; 24 merged changes in Stet-AI/stet. Compare the individual calls below to choose which setup to investigate.
Where do these coding results come from?: Both coding agents worked the same 24 merged changes from the private Stet-AI/stet repository, each in the real repo as it was before the change. The repo’s tests and judges graded every patch against the merged change; no public benchmark scores are used.
Are reasoning settings combined into one model score?: No. Each effort pairing keeps its own recorded quality call, head-to-head record, cost and agent time. Incompatible scores are never averaged.