Introducing Kimi K3.1 Pro
Moonshot AI's launch post for Kimi K3.1 Pro, a 3T-class model with a 1-million-token context window and a faster Highspeed variant, with benchmark results and an account of removing cyberattack-related training data before the open-weights release.
Summary of Moonshot AI's launch post (October 2026). Original: kimi.ai/blog/kimi-k3-1-pro
The model
Moonshot AI presents Kimi K3.1 Pro as its most capable 3T-class open-weight model so far, with a 1-million-token context window. A companion model, Kimi K3.1 Pro Highspeed, runs about two to three times faster.
Availability
Live on Kimi.ai, Kimi Work, Kimi Code and the Kimi API (model IDs
kimi-k3.1-proandkimi-k3.1-pro-highspeed)Full model weights promised by November 4, 2026
Enterprise access is off by default
Moonshot's reported benchmarks
Benchmark
Kimi K3.1 Pro
Comparison
Terminal-Bench 4.0
65.7%
Claude Opus 5.5 66.4%, GPT-6 Astra 58.2%
Kimi Code Bench 3.0 (long-horizon)
61.2%
Kimi K3 55.95%
BenchCAD (with tools)
92.7%
Claude Opus 5.5 96.2%, GPT-6 Astra 95.9%
OfficeQA Pro v2
64.4
described as state of the art
Frontier Finance
58.6
Internal finance modeling
57.6
Kimi K3 50.4
Internal financial research
70.4
Kimi K3 58.8
Internal business operations
37
Kimi K3 8.9
Kimi Hallu Bench Hard (with tools, lower is better)
8.3%
Claude Opus 5.5 15.0%, GPT-6 Astra 5.5%
Results were run at maximum reasoning effort, temperature 1.0 and top-p 0.95.
Removing cyber capability
Moonshot says it identified and removed cyber-related data from both training stages, and that without this step the model's downstream cyber capability would have risen substantially. On its internal xdev-bench-hard exploit test (pass@3):
Kimi K3.1 Pro without the removal: 90.5%
Kimi K3.1 Pro as released: 81.0%
Kimi K3: 33.3%
GLM-5.3: 57.1%, DeepSeek-V4.1-Flash: 85.7%, GPT-5.6 Sol: 100
Moonshot argues the released model's cyber capability remains behind frontier proprietary models that have been publicly available for at least three months.