Claude Opus 5 vs GPT-5.6 Luna
Claude Opus 5 leads GPT-5.6 Luna by 14.0 points on the Shogo Agent Index, but their ranges overlap, so treat it as close. Claude Opus 5 wins 22 and GPT-5.6 Luna wins 0 of 23 shared benchmarks. GPT-5.6 Luna is cheaper for agent workloads.
Add more models to this comparison →Pillars
| Pillar | Claude Opus 5 | GPT-5.6 Luna |
|---|---|---|
| Agentic coding | 74 | 59 |
| Long-horizon autonomy | 74 | 61 |
| Knowledge-work agents | 52 | 39 |
| Computer use | 56 | 40 |
| Security & context learning | 59 | 51 |
Shared benchmarks
| Benchmark | Claude Opus 5 | GPT-5.6 Luna |
|---|---|---|
| APEX-Agentsindex | 65.8% | 43.0% |
| DeepSWEindex | 73.6% | 67.2% |
| Vending-Bench 2index | $11,182 | $4,095 |
| CursorBenchindex | 46.6% | 35.9% |
| ARC-AGI | 97.5% | 88.0% |
| ARC-AGI-2 | 90.4% | 59.5% |
| BALROG | 63.4% | 45.6% |
| Chess Puzzles | 43.0% | 43.0% |
| CritPt | 29.1% | 20.6% |
| DTBench | 97.9% | 89.1% |
| FrontierCode | 53.4% | 39.8% |
| FrontierMath Tier 4 | 73.2% | 61.0% |
| FrontierMath Tiers 1-3 | 85.6% | 82.1% |
| GPQA Diamond | 93.9% | 91.6% |
| LMCA | 64.5% | 48.5% |
| MirrorCode | 87.8% | 21.1% |
| OTIS Mock AIME | 98.9% | 98.3% |
| ProofBench | 99.0% | 60.0% |
| SciCode | 56.4% | 53.6% |
| SimpleBench | 80.6% | 46.8% |
| SimpleQA Verified | 59.9% | 41.0% |
| WebDev Arena | 1687 | 1519 |
| WeirdML v3 | 19.0% | 5.2% |
Specs and price
| Claude Opus 5 | GPT-5.6 Luna | |
|---|---|---|
| Agent Index | 65.4 (#11) | 51.4 (#34) |
| Agent $/M | $2.31 | $0.10 |
| Input / output $/M | $5.00 / $25.00 | $0.20 / $1.20 |
| Context | 1M | 1.1M |
| Released | Jul 24, 2026 | Jul 9, 2026 |
| Weights | Closed | Closed |
| API hosts | 6 | 5 |
Don’t pick one. Route.
Shogo can run Claude Opus 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