News - Kimi K3.1 Pro — Open-Weight AI Approaches the Proprietary Frontier
Chinese AI company Moonshot AI has introduced Kimi K3.1 Pro, a new reasoning model that comes remarkably close to the best proprietary models from Anthropic and OpenAI.
According to independent evaluations by Artificial Analysis, Kimi K3.1 Pro scores 55 points on the Intelligence Index, compared with 58 for Claude Opus 5.5. Particularly impressive is its performance on Terminal-Bench 4.0, where it achieves 67%, up from just 13% for the previous Kimi K3 model.
Key technical facts
- Architecture: Mixture-of-Experts (MoE)
- Parameters: 2.8 trillion total, approximately 104 billion active
- Context window: 1 million tokens
- Capabilities: Reasoning, coding, multimodal understanding and agentic workflows
- API pricing: $2 per million input tokens / $10 per million output tokens
- Model weights: Moonshot plans to release them by November 4, 2026
The last point is important: although Moonshot describes K3.1 Pro as an open-weight model, the weights are not yet publicly available.
Why does it matter?
Open-weight models are rapidly approaching proprietary frontier-model capabilities. This creates new opportunities for organizations requiring local deployment, data sovereignty, infrastructure control and reduced vendor lock-in.
However, parameter count and benchmark scores are only part of the story. Actual deployment cost, GPU requirements, inference performance and reliability still need independent evaluation.
Summary: The gap between open-weight and proprietary AI is narrowing, but the real competition will be about the cost of delivering reliable intelligence in production.