Kimi K3 vs GPT-5.6 Luna: Benchmark Comparison
Kimi K3 has the higher score on 18 of 22 shared benchmarks; GPT-5.6 Luna leads on 4.
The largest observed score gap is 27.00 pts on ProofBench v1.1 , where Kimi K3 leads.
Reported ±1 standard-error ranges overlap on 9 of 22 shared benchmarks with comparable uncertainty data. This is not a pairwise statistical significance test.
Shared benchmark results
Scores and reported standard errors come from the latest visible Vals benchmark version available for both models.
| Benchmark | Kimi K3 | GPT-5.6 Luna | Gap | Reported uncertainty |
|---|---|---|---|---|
| Legal Research Bench | 46.15% ±3.46 | 36.54% ±3.35 | 9.62 pts | Reported ±1 SE ranges do not overlap |
| LegalBench | 86.21% ±0.41 | 84.03% ±0.42 | 2.18 pts | Reported ±1 SE ranges do not overlap |
| MortgageTax | 66.34% ±0.91 | 67.29% ±0.92 | 0.95 pts | Reported ±1 SE ranges overlap |
| Tax Agent Bench | 33.37% ±3.39 | 24.82% ±3.11 | 8.55 pts | Reported ±1 SE ranges do not overlap |
| TaxEval v2 | 75.72% ±0.84 | 76.17% ±0.84 | 0.45 pts | Reported ±1 SE ranges overlap |
| MedCode | 49.36% ±2.20 | 42.39% ±2.27 | 6.97 pts | Reported ±1 SE ranges do not overlap |
| MedScribe | 88.05% ±1.98 | 84.39% ±2.58 | 3.66 pts | Reported ±1 SE ranges overlap |
| ProofBench v1.1 | 87.00% ±3.38 | 60.00% ±4.92 | 27.00 pts | Reported ±1 SE ranges do not overlap |
| BioMysteryBench | 72.59% ±1.61 | 61.48% ±0.74 | 11.11 pts | Reported ±1 SE ranges do not overlap |
| Terminal-Bench-Science | 1.43% ±1.43 | 0.00% ±0.00 | 1.43 pts | Reported ±1 SE ranges overlap |
| GPQA Diamond | 92.93% ±1.31 | 91.67% ±1.74 | 1.26 pts | Reported ±1 SE ranges overlap |
| MMLU Pro | 87.97% ±0.32 | 86.04% ±0.35 | 1.94 pts | Reported ±1 SE ranges do not overlap |
| MMMU Pro | 88.15% ±0.78 | 85.03% ±0.86 | 3.12 pts | Reported ±1 SE ranges do not overlap |
| SAGE | 52.78% ±3.41 | 44.22% ±3.33 | 8.56 pts | Reported ±1 SE ranges do not overlap |
| IOI | 48.94% ±9.82 | 61.78% ±11.61 | 12.83 pts | Reported ±1 SE ranges overlap |
| ProgramBench | 2.00% ±0.99 | 0.00% ±0.00 | 2.00 pts | Reported ±1 SE ranges do not overlap |
| SWE-bench | 93.40% ±1.11 | 93.00% ±1.14 | 0.40 pts | Reported ±1 SE ranges overlap |
| Terminal-Bench 4.0 | 17.17% ±0.51 | 11.62% ±1.01 | 5.56 pts | Reported ±1 SE ranges do not overlap |
| Vibe Code Bench 1-100 | 18.24% ±3.92 | 22.59% ±4.00 | 4.35 pts | Reported ±1 SE ranges overlap |
| Vibe Code Bench v1.1 | 84.97% ±2.75 | 77.06% ±3.07 | 7.91 pts | Reported ±1 SE ranges do not overlap |
| CyberBench v1.1 | 75.24% ±5.56 | 73.63% ±5.56 | 1.61 pts | Reported ±1 SE ranges overlap |
| Public Benefits Bench v1.1 | 68.20% ±1.21 | 61.16% ±1.27 | 7.04 pts | Reported ±1 SE ranges do not overlap |
Performance by category
| Category | Kimi K3 average | GPT-5.6 Luna average |
|---|---|---|
| Legal | 66.18% | 60.29% |
| Finance | 58.47% | 56.09% |
| Healthcare | 68.70% | 63.39% |
| Math | 87.00% | 60.00% |
| Science | 37.01% | 30.74% |
| Academic | 89.68% | 87.58% |
| Education | 52.78% | 44.22% |
| Coding | 44.12% | 44.34% |
| Cyber | 75.24% | 73.63% |
| Social Mobility | 68.20% | 61.16% |
Cost and latency
Cost per test appears only where the benchmark reports it for both models. Latency is the measured completion time for that benchmark.
| Benchmark | Kimi K3 cost | GPT-5.6 Luna cost | Kimi K3 latency | GPT-5.6 Luna latency |
|---|---|---|---|---|
| Legal Research Bench | $3.47 | $0.85 | 12m40s | 39m34s |
| LegalBench | N/A | N/A | 7.11s | 5.89s |
| MortgageTax | N/A | N/A | 78.94s | 22.84s |
| Tax Agent Bench | $2.79 | $0.47 | 40m15s | 42m46s |
| TaxEval v2 | N/A | N/A | 2m25s | 76.52s |
| MedCode | N/A | N/A | 36.99s | 81.28s |
| MedScribe | N/A | N/A | 38.24s | 116.89s |
| ProofBench v1.1 | $1.65 | $0.12 | 34m31s | 8m34s |
| BioMysteryBench | $1.06 | $0.09 | 4m55s | 18m31s |
| Terminal-Bench-Science | $21.30 | $0.58 | 4h49m | 2h03m |
| GPQA Diamond | N/A | N/A | 2m02s | 53.52s |
| MMLU Pro | N/A | N/A | 38.72s | 16.95s |
| MMMU Pro | N/A | N/A | 106.52s | 35.10s |
| SAGE | N/A | N/A | 47.30s | 51.88s |
| IOI | $14.17 | $0.58 | 4h18m | 1h10m |
| ProgramBench | $70.48 | $0.94 | 5h42m | 41m34s |
| SWE-bench | $0.76 | $0.04 | 10m19s | 3m21s |
| Terminal-Bench 4.0 | $12.48 | $0.73 | 3h09m | 35m55s |
| Vibe Code Bench 1-100 | $9.40 | $4.91 | 1h48m | 2h03m |
| Vibe Code Bench v1.1 | $10.01 | $0.73 | 16m39s | 25m52s |
| CyberBench v1.1 | $2.13 | $0.40 | 30m44s | 16m15s |
| Public Benefits Bench v1.1 | $0.91 | $0.27 | 7m19s | 38m40s |
Results available only for Kimi K3
- Vals Finance Index
- Vals Index
- Vals Inference Provider Index
- Vals Legal Index
- Vals RSI Index
- HLAB-Vals-1.1
- EMB v1.1
- Finance Agent v2.1
- Code Migration v1.1
- LiveCodeBench
- SWE-bench Multimodal
- Time Horizon Index: KSP
Results available only for GPT-5.6 Luna
- Vals Science Index
- MysteryMechanism
- SkillsBench
- SRE Bench