Specifications
- Lab
- Moonshot AI (China)
- Weights
- Open weights
- Context window
- 1M tokens
- Maximum output
- 944k tokens
- Input / output price
- $2.7 / $13.5 per million tokens
- Input modalities
- Text
- Image
- Video
- Reasoning
- Yes
- OpenRouter identifier
- moonshotai/kimi-k3
At a glance
Updated 5 October 2026
Business Index
99.8%± 0.7 (95% margin of error)
Rank 6 / 27
Average cost per test
$0.021
22nd cheapest of 27
Median time
7.0 s
17th fastest of 27
- Average hallucinations
- 1.0%
- Benchmarks taken
- 1 / 1
Scores by business function
The model's mean accuracy across the benchmarks of each business function.
Finance
Not tested
Accounting
99.8%
Human resources
Not tested
Legal
Not tested
Sales
Not tested
Marketing
Not tested
Customer service
Not tested
Procurement & logistics
Not tested
IT
Not tested
Management
Not tested
Results by benchmark
Accounting
- Advertising invoice extraction
99.8%± 0.7
rank 6 of 27
Bars on a 0 to 100% scale. The rank compares the models tested on the same benchmark.
Stability from run to run
A given model alias can change behaviour without notice; re-testing it at every run is how that drift gets caught.
Mean of the model's accuracy across the benchmarks it appears in, run after run. The vertical axis is truncated so that small gaps stay readable.
View the values
| Run | Mean accuracy | Benchmarks |
|---|---|---|
| 2 October 2026 | 99.6% | 1 |
| 4 October 2026 | 99.6% | 1 |
| 5 October 2026 | 99.8% | 1 |