Kimi K3.1 Pro Finally Launches: The Most Powerful Open-Source Cybersecurity Model

Beating AI Express: MoonShot releases Kimi K3.1 Pro, the first Kimi model with "Pro" in its name. It retains K3’s 2.8 trillion parameter base, primarily enhancing agent capabilities via large-scale reinforcement learning. In third-party firm Artificial Analysis’s cybersecurity defense index, K3.1 Pro scored 64% – 23 percentage points higher than K3, surpassing DeepSeek V4.1 Flash’s 57% to rank first among open-source models. Only OpenAI’s non-public GPT-6 Sol (Daybreak Blue) scored higher on the entire list, at 69%. Comprehensive capabilities also improved significantly: Artificial Analysis rated it 55 points, an 11-point jump from K3, placing it 4th globally, just 3 points behind top-ranked Claude Opus 5.5. In Terminal-Bench 4.0, its terminal programming task accuracy surged from K3’s 13% to 67%, securing the top spot. 3D capabilities showed mixed results: it scored 92.7% on official BenchCAD, close to Opus 5.5’s 96.2%; however, its early performance on independent interactive 3D benchmark AA-3D was not as strong as in programming tasks. K3.1 Pro’s API token unit price is one-third lower than K3’s, but Artificial Analysis estimates its average per-question cost for cybersecurity defense evaluations rose from $2.45 to $2.81, an ~15% increase. The main driver is input token consumption nearly doubling, which offset the price cut. The model is now live on Kimi, Work, Code, and API; full weights are scheduled to be released no later than November 4. MoonShot also proactively removed some cybersecurity attack/defense training data to reduce abuse risks post full-weight release. In internal official exploit tests, the control version without risk mitigation measures had a 90.5% success rate, while the final version hit 81%.