AI developer Moonshot AI introduced its flagship model Kimi K3.1 Pro alongside an accelerated API endpoint on October 23, 2026. The new system relies on a Mixture-of-Experts architecture spanning a total of 2.8 trillion parameters, with 104 billion parameters activated during any single inference pass. In addition to the standard model, the company debuted a Highspeed API capable of executing queries two to three times faster than prior endpoints. This performance upgrade specifically addresses latency-sensitive workloads that demand rapid turnarounds across complex analytical pipelines.
A standout technical capability of Kimi K3.1 Pro is its context window of up to one million tokens. Moonshot documented leading benchmark performances in processing structured financial records and complex spreadsheets, specifically citing high marks on GDP.pdf and OfficeQA Pro. The model also demonstrated competitive execution within autonomous operational frameworks. On Terminal-Bench 4.0, a benchmark that tests agentic capabilities across command-line interfaces and multi-step system workflows, Kimi K3.1 Pro achieved a success rate of 65.7 percent.
The pricing structure positions Moonshot AI directly against legacy enterprise providers in the frontier AI market. Usage of the API is set at two US dollars per million input tokens and ten US dollars per million output tokens. To win over risk-averse institutional clients, including investment banks, hedge funds, and quantitative research firms, Moonshot rolled out a binding zero data retention option. This policy guarantees that confidential financial spreadsheets, customer balance sheets, and proprietary trading rules are never retained or recycled into future model updates.
Ahead of a broader release of the model weights scheduled for early November 2026, Moonshot AI undertook targeted safety interventions during dataset curation. The company systematically removed offensive cyber data and exploit patterns from the training corpus to prevent regulatory pushback regarding automated security vulnerabilities. By sanitizing the foundation weights before distribution, the lab aims to ensure the architecture cannot be easily weaponized for automated cyber reconnaissance. This initiative aligns with growing international regulatory pressure to mitigate dual-use capabilities early in the model development cycle.
The timing of the launch aligns with widespread security concerns across enterprise and financial IT departments. In the wake of recent automated vulnerability discoveries, vendors must prove that multi-agent systems do not create unmanageable supply chain or operational risks. Moonshot's dual focus on document comprehension, agent workflows, and sanitized model training illustrates an evolving industry consensus. Enterprise buyers in highly regulated industries are increasingly evaluating models not merely on reasoning benchmarks, but on auditable compliance and deterministic safety constraints.