GPT-5.6 Luna vs Kimi K3
Kimi K3 leads GPT-5.6 Luna by 3.5 points on the Shogo Agent Index, but their ranges overlap, so treat it as close. GPT-5.6 Luna wins 4 and Kimi K3 wins 17 of 21 shared benchmarks. GPT-5.6 Luna is cheaper for agent workloads.
Add more models to this comparison →Pillars
| Pillar | GPT-5.6 Luna | Kimi K3 |
|---|---|---|
| Agentic coding | 59 | 63 |
| Long-horizon autonomy | 61 | 65 |
| Knowledge-work agents | 39 | 42 |
| Computer use | 40 | 43 |
| Security & context learning | 51 | 53 |
Shared benchmarks
| Benchmark | GPT-5.6 Luna | Kimi K3 |
|---|---|---|
| APEX-Agentsindex | 43.0% | 47.8% |
| DeepSWEindex | 67.2% | 68.5% |
| Vending-Bench 2index | $4,095 | $5,165 |
| ARC-AGI | 88.0% | 94.5% |
| ARC-AGI-2 | 59.5% | 60.4% |
| Chess Puzzles | 43.0% | 37.0% |
| CritPt | 20.6% | 23.4% |
| DTBench | 89.1% | 91.2% |
| FrontierCode | 39.8% | 44.2% |
| FrontierMath Tier 4 | 61.0% | 39.0% |
| FrontierMath Tiers 1-3 | 82.1% | 72.2% |
| GPQA Diamond | 91.6% | 93.1% |
| LMCA | 48.5% | 52.7% |
| MirrorCode | 21.1% | 54.4% |
| OTIS Mock AIME | 98.3% | 97.2% |
| ProofBench | 60.0% | 87.0% |
| SciCode | 53.6% | 59.5% |
| SimpleBench | 46.8% | 60.7% |
| SimpleQA Verified | 41.0% | 50.6% |
| WebDev Arena | 1519 | 1674 |
| WeirdML v3 | 5.2% | 7.5% |
Specs and price
| GPT-5.6 Luna | Kimi K3 | |
|---|---|---|
| Agent Index | 51.4 (#34) | 54.9 (#26) |
| Agent $/M | $0.10 | $1.39 |
| Input / output $/M | $0.20 / $1.20 | $3.00 / $15.00 |
| Context | 1.1M | 1M |
| Released | Jul 9, 2026 | Jul 16, 2026 |
| Weights | Closed | Open |
| API hosts | 5 | 29 |
Don’t pick one. Route.
Shogo can run GPT-5.6 Luna and Kimi K3 side by side and send each step of an agent to the one that’s cheaper for it.
Try both in Shogo