Skip to content
VECTOR WIREAI INTELLIGENCE
UTC

GPT-5.2 vs GPT-5.6 Luna

45 SHARED BENCHMARKS

Across 45 shared benchmarks, GPT-5.2 scores higher on 14 and GPT-5.6 Luna on 31. The widest gap is τ-Bench V3 · Banking, where GPT-5.6 Luna scores 31.1 against 11.1. GPT-5.6 Luna is the cheaper of the two on tracked API pricing ($0.20 against $1.75 per million input tokens).

OPENAIVSOPENAI45 SHARED14–31 HEAD-TO-HEADNEWEST SCORE ADDED

At a glance

MakerOpenAIOpenAI
Released
Price per 1M tokens input / output$1.75 / $14.00$0.20 / $1.20
Cost of 1M in + 1M out$15.75$1.40 11.3× less
Head-to-head of 45 shared benchmarks14 wins31 wins
Scores tracked independently verified204 64 ◆201 43 ◆

Release dates per Artificial Analysis. Prices: Artificial Analysis for GPT-5.2; Artificial Analysis for GPT-5.6 Luna. ◆ = independently verified score.

Where each leadsby capability area · benchmark wins

AreaGPT-5.2WINSGPT-5.6 Luna
Reasoning06GPT-5.6 Luna leads 6 of 6 · widest: CritPt 11.6 vs 20.6
Coding15GPT-5.6 Luna leads 5 of 6 · widest: SWE-bench Pro 29.9 vs 62.7
Agentic13GPT-5.6 Luna leads 3 of 4 · widest: τ-Bench V3 · Banking 11.1 vs 31.1
Factuality21GPT-5.2 leads 2 of 3 · widest: AA-Omniscience · Non-hallucination 18.8 vs 7.4
Instruction Following10GPT-5.2 leads 1 of 1
Long Context11Even, 1–1 of 2
Math23GPT-5.6 Luna leads 3 of 5 · widest: FrontierMath Tier 4 31.7 vs 58.5
Multimodal03GPT-5.6 Luna leads 3 of 3 · widest: MMMU-Pro 74.6 vs 78.6

Biggest gaps

GPT-5.2 pulls furthest ahead on

  1. AA-Omniscience · Non-hallucination18.8 vs 7.4
  2. MRCR v2 8 needle 128k (average)83.8 vs 74.8
  3. LiveBench · Mathematics93.2 vs 87.2

GPT-5.6 Luna pulls furthest ahead on

  1. τ-Bench V3 · Banking31.1 vs 11.1
  2. SWE-bench Pro62.7 vs 29.9
  3. FrontierMath Tier 458.5 vs 31.7

Every shared benchmark45 · grouped by area

Reasoning 6

BenchmarkGPT-5.2MARGINGPT-5.6 Luna
ARC-AGI-252.9◆ 59.6
CritPt11.620.6
GPQA Diamond90.391.1
LiveBench · Reasoning83.2 ◆◆ 85.6
SimpleBench45.8 ◆◆ 46.8

Coding 6

BenchmarkGPT-5.2MARGINGPT-5.6 Luna
LMArena · WebDev1415 ◆◆ 1520
LiveBench · Coding76.1 ◆◆ 82.9
SciCode52.153.6
SWE-bench Pro29.9 ◆62.7

Agentic 4

Instruction Following 1

Long Context 2

Math 5

BenchmarkGPT-5.2MARGINGPT-5.6 Luna
AIME 202698.3 ◆97.6
HMMT Feb. 202697 ◆98.5
LiveBench · Mathematics93.2 ◆◆ 87.2

Multimodal 3

BenchmarkGPT-5.2MARGINGPT-5.6 Luna
LMArena · Vision1246 ◆◆ 1259
CharXiv (RQ)82.182.7
MMMU-Pro74.678.6

Each score is the one the model's own page shows — the most authoritative tracked result for that benchmark, on the benchmark's standard methodology (source-attributed). ↓ marks lower-is-better metrics; ◆ an independently verified score. Open either model for its full surface, provenance and pricing. Quoted rates are the price-setter row we currently track for each model — its direct or vendor-official listing where one exists (direct, direct), otherwise the lowest tracked offer.

Compare GPT-5.2 withALL PAIRINGS →

Compare GPT-5.6 Luna withALL PAIRINGS →