GPT-5.5 vs GPT-5.6 Luna
GPT-5.5 leads GPT-5.6 Luna by 5.3 points on the Shogo Agent Index, but their ranges overlap, so treat it as close. GPT-5.5 wins 17 and GPT-5.6 Luna wins 3 of 20 shared benchmarks. GPT-5.6 Luna is cheaper for agent workloads.
Add more models to this comparison →Pillars
| Pillar | GPT-5.5 | GPT-5.6 Luna |
|---|---|---|
| Agentic coding | 63 | 59 |
| Long-horizon autonomy | 67 | 61 |
| Knowledge-work agents | 46 | 39 |
| Computer use | 43 | 40 |
| Security & context learning | 55 | 51 |
Shared benchmarks
| Benchmark | GPT-5.5 | GPT-5.6 Luna |
|---|---|---|
| APEX-Agentsindex | 55.1% | 43.0% |
| DeepSWEindex | 67.0% | 67.2% |
| Vending-Bench 2index | $7,524 | $4,095 |
| ARC-AGI | 95.0% | 88.0% |
| ARC-AGI-2 | 85.0% | 59.5% |
| Chess Puzzles | 63.0% | 43.0% |
| CritPt | 27.1% | 20.6% |
| DTBench | 96.0% | 89.1% |
| FrontierCode | 43.0% | 39.8% |
| FrontierMath Tier 4 | 72.5% | 61.0% |
| FrontierMath Tiers 1-3 | 85.3% | 82.1% |
| GPQA Diamond | 94.0% | 91.6% |
| LMCA | 54.3% | 48.5% |
| MirrorCode | 23.3% | 21.1% |
| OTIS Mock AIME | 100.0% | 98.3% |
| ProofBench | 50.0% | 60.0% |
| SciCode | 56.1% | 53.6% |
| SimpleBench | 69.0% | 46.8% |
| SimpleQA Verified | 63.0% | 41.0% |
| WebDev Arena | 1510 | 1519 |
Specs and price
| GPT-5.5 | GPT-5.6 Luna | |
|---|---|---|
| Agent Index | 56.7 (#19) | 51.4 (#34) |
| Agent $/M | $2.55 | $0.10 |
| Input / output $/M | $5.00 / $30.00 | $0.20 / $1.20 |
| Context | 1.1M | 1.1M |
| Released | Apr 23, 2026 | Jul 9, 2026 |
| Weights | Closed | Closed |
| API hosts | 4 | 5 |
Don’t pick one. Route.
Shogo can run GPT-5.5 and GPT-5.6 Luna side by side and send each step of an agent to the one that’s cheaper for it.
Try both in Shogo