GPT-5.6 Luna vs GPT-5.6 Sol
GPT-5.6 Sol leads GPT-5.6 Luna by 10.4 points on the Shogo Agent Index, but their ranges overlap, so treat it as close. GPT-5.6 Luna wins 1 and GPT-5.6 Sol wins 22 of 23 shared benchmarks. GPT-5.6 Luna is cheaper for agent workloads.
Add more models to this comparison →Pillars
| Pillar | GPT-5.6 Luna | GPT-5.6 Sol |
|---|---|---|
| Agentic coding | 59 | 71 |
| Long-horizon autonomy | 61 | 72 |
| Knowledge-work agents | 39 | 46 |
| Computer use | 40 | 53 |
| Security & context learning | 51 | 58 |
Shared benchmarks
| Benchmark | GPT-5.6 Luna | GPT-5.6 Sol |
|---|---|---|
| APEX-Agentsindex | 43.0% | 51.4% |
| DeepSWEindex | 67.2% | 72.7% |
| Vending-Bench 2index | $4,095 | $9,619 |
| CursorBenchindex | 35.9% | 41.7% |
| ARC-AGI | 88.0% | 97.5% |
| ARC-AGI-2 | 59.5% | 92.5% |
| BALROG | 45.6% | 60.0% |
| Chess Puzzles | 43.0% | 68.0% |
| CritPt | 20.6% | 32.3% |
| DTBench | 89.1% | 96.0% |
| FrontierCode | 39.8% | 47.5% |
| FrontierMath Tier 4 | 61.0% | 82.9% |
| FrontierMath Tiers 1-3 | 82.1% | 89.1% |
| GPQA Diamond | 91.6% | 93.5% |
| LMCA | 48.5% | 59.2% |
| MirrorCode | 21.1% | 20.0% |
| OTIS Mock AIME | 98.3% | 100.0% |
| ProofBench | 60.0% | 83.0% |
| SciCode | 53.6% | 57.8% |
| SimpleBench | 46.8% | 71.7% |
| SimpleQA Verified | 41.0% | 69.7% |
| WebDev Arena | 1519 | 1617 |
| WeirdML v3 | 5.2% | 15.0% |
Specs and price
| GPT-5.6 Luna | GPT-5.6 Sol | |
|---|---|---|
| Agent Index | 51.4 (#34) | 61.8 (#14) |
| Agent $/M | $0.10 | $1.85 |
| Input / output $/M | $0.20 / $1.20 | $4.00 / $20.00 |
| Context | 1.1M | 1.1M |
| Released | Jul 9, 2026 | Aug 7, 2026 |
| Weights | Closed | Closed |
| API hosts | 5 | 5 |
Don’t pick one. Route.
Shogo can run GPT-5.6 Luna and GPT-5.6 Sol side by side and send each step of an agent to the one that’s cheaper for it.
Try both in Shogo