Models/gpt-5-6-luna

GPT 5.6 Luna

About 17th on The Slop Index after inversion (100 − Slop = 65). Fifth on Android Bench 1.0 at 87.6%, run cost $7.2.

OpenAI · Proprietary · Fetched 2026-10-10

1519

#59 / 153 · ±6

1517

#39 / 65 · ±10

17.3%

#30 / 35 · ±2.8

Benchmark scores

Boards older than 90 days are not counted as recommendations

BenchmarkSettingScoreRankCostAs of

LMArena WebDevxhigh1519±6#59—2026-10-27

LMArena Image-to-WebDevxhigh1517±10#39—2026-10-19

Terminal-Bench 4.0max · Codex17.3±2.8#30$346.672026-10-06

The Slop Index—63#19—2026-10-28

Convex Evals2 runs63.39#26$0.200922026-10-28

ReactBenchmax33.830.5–37.2#32$0.842026-10-19

DeepSWE v1.1max · mini-swe-agent67.263.2–71.2#10$3.032026-09-22

SWE-bench Verified—This board has stopped updating——2026-02-26

Aider Polyglot—This board has stopped updating——2025-11-20

Why

On Android standard tasks, about four points behind first at roughly a twenty-fourth of the cost.

When not

Haiku 5.5 now leads Slop Index; do not still call this the least-slop model.
The cheap long-horizon row in this family is GPT 6 Luna at 10.7%. Do not give multi-module work to the cheap tier alone.

Collected from scenario boards

ModelWebDevPrice $/M