Skip to the comparison

DeepSeek V4.1 Flash (10/17) max vs GPT 5.6 Luna low

GPT 5.6 Luna low leads on speed and cost; DeepSeek V4.1 Flash (10/17) max has a slight edge on quality.

Quality
+4.9 pts
DeepSeek V4.1 Flash (10/17) max
Slight edge
3–1 head to head, 20 tied
Speed
5.6× faster
GPT 5.6 Luna low
6.7 min vs 37.2 min per task
Cost
5.5× cheaper
GPT 5.6 Luna low
$0.05 vs $0.28 per task

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

DeepSeek V4.1 Flash (10/17) vs GPT 5.6 Luna: all effort pairings

24 tasksData from

Stet score

DeepSeek V4.1 Flash (10/17) maxcodex · max reasoningRank 11 of 18

48.395% range 42.9–53.44.9 points higher

GPT 5.6 Luna lowcodex · low reasoningRank 17 of 18

43.495% range 36.6–49.9

A directional read: their 95% score ranges overlap.

Quality

Head-to-head matchups

DeepSeek V4.1 Flash (10/17) max won3

Tied20

GPT 5.6 Luna low won1

DeepSeek V4.1 Flash (10/17) max’s win share 55%, 95% range 48–63%Ties are within judge noise.

Task by task

DeepSeek V4.1 Flash (10/17) max solved 2 tasks that GPT 5.6 Luna low didn’t, and GPT 5.6 Luna low solved 1 that DeepSeek V4.1 Flash (10/17) max didn’t. Both solved 8; neither solved 11.

On the 8 tasks both solved, judges preferred DeepSeek V4.1 Flash (10/17) max’s patch on 1 and called 7 even.

2Only DeepSeek V4.1 Flash (10/17) max solved

  • 2 changed lines in 1 filestet-2450ca2dJudges preferred DeepSeek V4.1 Flash (10/17) max’s patch
  • 941 changed lines in 7 filesstet-commit-bbbbae09Judges preferred DeepSeek V4.1 Flash (10/17) max’s patch

1Only GPT 5.6 Luna low solved

  • 113 changed lines in 4 filesstet-commit-f0caded9

8Both solved

  • 4 changed lines in 1 filestet-43493c52
  • 6 changed lines in 1 filestet-3169aefa
  • 9 changed lines in 1 filestet-89dfbc27
  • 10 changed lines in 1 filestet-232b50ba
  • 23 changed lines in 2 filesstet-5276aa3f
  • 29 changed lines in 1 filestet-b28f4fe7
  • 84 changed lines in 1 filestet-66762b4cJudges preferred DeepSeek V4.1 Flash (10/17) max’s patch
  • 218 changed lines in 2 filesstet-commit-5d766dde

11Neither solved

  • 149 changed lines in 2 filesstet-commit-e928166f
  • 476 changed lines in 17 filesstet-commit-30506f8e
  • 481 changed lines in 3 filesstet-commit-5e075641
  • 552 changed lines in 11 filesstet-commit-60a0e6c2
  • 564 changed lines in 10 filesstet-commit-a8716b88
  • 640 changed lines in 8 filesstet-commit-342d0ac0
  • 763 changed lines in 8 filesstet-commit-87814583
  • 977 changed lines in 5 filesstet-commit-e41c2e2d
  • 1,236 changed lines in 16 filesstet-commit-d2ab0c25
  • 1,312 changed lines in 9 filesstet-commit-f8ae1177
  • 1,420 changed lines in 19 filesstet-commit-2826765e

Bar: changed lines in the merged change, log scale

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. 2 of the 24 tasks are left out because one of the two has no scored result on them.

All 22 compared tasks
Task by task
TaskDeepSeek V4.1 Flash (10/17) maxGPT 5.6 Luna lowSolved byBetter patch
stet-3169aefa · 6 changed lines in 1 filePassed tests$0.06 · 15.9 minPassed tests$0.02 · 4.9 minBothEven
stet-89dfbc27 · 9 changed lines in 1 filePassed tests$0.05 · 4.9 minPassed tests$0.04 · 6.5 minBothEven
stet-b28f4fe7 · 29 changed lines in 1 filePassed tests$0.06 · 28.7 minPassed tests$0.03 · 2.6 minBothEven
stet-232b50ba · 10 changed lines in 1 filePassed tests$0.05 · 28.0 minPassed tests$0.04 · 14.5 minBothEven
stet-5276aa3f · 23 changed lines in 2 filesPassed tests$0.02 · 14.4 minPassed tests$0.05 · 13.4 minBothEven
stet-commit-bbbbae09 · 941 changed lines in 7 filesPassed tests$0.65 · 47.7 minFailed$0.04 · 6.6 minDeepSeek V4.1 Flash (10/17) max onlyDeepSeek V4.1 Flash (10/17) max
stet-66762b4c · 84 changed lines in 1 filePassed tests$0.08 · 9.1 minEquivalent to the merged fix$0.02 · 3.7 minBothDeepSeek V4.1 Flash (10/17) max
stet-commit-5d766dde · 218 changed lines in 2 filesEquivalent to the merged fix$0.09 · 18.5 minEquivalent to the merged fix$0.05 · 10.7 minBothEven
stet-43493c52 · 4 changed lines in 1 filePassed tests$0.14 · 16.0 minPassed tests$0.03 · 4.0 minBothEven
stet-commit-f0caded9 · 113 changed lines in 4 filesFailed$0.09 · 14.6 minEquivalent to the merged fix$0.04 · 2.9 minGPT 5.6 Luna low onlyEven
stet-commit-f8ae1177 · 1,312 changed lines in 9 filesFailed$0.54 · 75.0 minFailed$0.05 · 6.7 minNeitherEven
stet-2450ca2d · 2 changed lines in 1 filePassed tests$0.07 · 10.1 minFailed$0.02 · 4.2 minDeepSeek V4.1 Flash (10/17) max onlyDeepSeek V4.1 Flash (10/17) max
stet-commit-342d0ac0 · 640 changed lines in 8 filesFailed$0.20 · 38.3 minFailed$0.12 · 12.4 minNeitherEven
stet-commit-60a0e6c2 · 552 changed lines in 11 filesFailed$0.34 · 39.2 minFailed$0.04 · 2.6 minNeitherEven
stet-commit-5e075641 · 481 changed lines in 3 filesFailed$0.46 · 53.1 minFailed$0.07 · 5.7 minNeitherEven
stet-commit-2826765e · 1,420 changed lines in 19 filesFailed$0.56 · 50.2 minFailed$0.04 · 7.2 minNeitherEven
stet-commit-a8716b88 · 564 changed lines in 10 filesFailed$0.30 · 40.2 minFailed$0.10 · 5.8 minNeitherEven
stet-commit-e928166f · 149 changed lines in 2 filesFailed$0.29 · 42.0 minFailed$0.06 · 3.4 minNeitherEven
stet-commit-30506f8e · 476 changed lines in 17 filesFailed$0.38 · 48.8 minFailed$0.12 · 10.1 minNeitherEven
stet-commit-87814583 · 763 changed lines in 8 filesFailed$0.50 · 62.2 minFailed$0.03 · 2.7 minNeitherEven
stet-commit-d2ab0c25 · 1,236 changed lines in 16 filesFailed$0.71 · 76.4 minFailed$0.05 · 5.0 minNeitherEven
stet-commit-e41c2e2d · 977 changed lines in 5 filesFailed$0.96 · 110.4 minFailed$0.10 · 8.3 minNeitherEven

Cost and time on each task

Each dot is one task, on a log scale: filled where the setup solved it, hollow where it didn’t. The line marks the median.

Cost per task

DeepSeek V4.1 Flash (10/17) maxmean $0.30 · median $0.25

GPT 5.6 Luna lowmean $0.05 · median $0.04

GPT 5.6 Luna low cost less on 21 of 22 tasks; DeepSeek V4.1 Flash (10/17) max on 1.

Agent time per task

DeepSeek V4.1 Flash (10/17) maxmean 38.3 min · median 38.7 min

GPT 5.6 Luna lowmean 6.5 min · median 5.7 min

GPT 5.6 Luna low finished faster on 21 of 22 tasks; DeepSeek V4.1 Flash (10/17) max on 1.

Where each setup is the better choice

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.

Small fixes

10 tasks · 2–84 changed lines in 1–2 files

DeepSeek V4.1 Flash (10/17) max solved 1 more, 8 of 8 to 7; GPT 5.6 Luna low cost 2.1× less per task.

GPT 5.6 Luna low was 2.4× faster.

Small fixes
DeepSeek V4.1 Flash (10/17) maxGPT 5.6 Luna low
Solved8/87/8
Cost / task$0.07$0.03
Agent time / task15.9 min6.7 min

Large, cross-cutting changes

Cheaper here without solving fewer: GPT 5.6 Luna low

14 tasks · 113–1,420 changed lines in up to 19 files

GPT 5.6 Luna low solved as many, 2 of 14, at 6.6× lower cost per task.

GPT 5.6 Luna low was 8.0× faster.

Large, cross-cutting changes
DeepSeek V4.1 Flash (10/17) maxGPT 5.6 Luna low
Solved2/142/14
Cost / task$0.43$0.07
Agent time / task51.2 min6.4 min

Side by side

Side by side
MetricDeepSeek V4.1 Flash (10/17) maxGPT 5.6 Luna low
Harnesscodexcodex
Reasoning effortmaxlow
Rank11 of 1817 of 18
Stet score48.395% range 42.9–53.443.495% range 36.6–49.9
Solved10/249 passed, 1 judged equivalent10/247 passed, 3 judged equivalent
Cost / task$0.28$0.05
Agent time / task37.2 min6.7 min

Questions

Which is better for coding, DeepSeek V4.1 Flash (10/17) max or GPT 5.6 Luna low?
DeepSeek V4.1 Flash (10/17) max has a slight quality edge, 3–1 head to head with 20 tied; GPT 5.6 Luna low leads on cost and agent time: 5.5× cheaper and 5.6× faster. Measured on 24 merged changes from Stet-AI/stet.
How were DeepSeek V4.1 Flash (10/17) max and GPT 5.6 Luna low tested?
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. Stet score: A setup’s average chance of beating another setup on a task, from 0 to 100; 50 is even with the field.

See which setup wins on your repo

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

Get started