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-k3Dated version, as of 8 October 2026: moonshotai/kimi-k3-20260715
At a glance
Updated 8 October 2026
Business Index
96.8%± 3.0 (95% margin of error)
Rank 17 / 27
54.8%100.0%
Average cost per test
$0.017
23rd cheapest of 27
$0.0002$0.058
Median time
5.8 s
17th fastest of 27
1.5 s29.1 s
- Average hallucinations
- 0.5%
- Benchmarks taken
- 2 / 2
Scores by business function
The model's mean accuracy across the benchmarks of each business function.
Finance
93.8%
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
Finance
- Financial report analysis
93.8%± 5.9
rank 17 of 27
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.
No benchmark has been published twice with this model yet: there is no curve to draw.