Moonshot AI Launches Kimi K3.1 Pro: 3 Trillion Parameter Model Outperforms GPT-6 Astra

Moonshot AI Introduces Kimi K3.1 Pro as Strongest Model Out of China

Moonshot AI has launched Kimi K3.1 Pro, a large language model with 3 trillion parameters that delivers competitive performance against Claude Opus 5.5 and GPT-6 Astra across multiple benchmarks1

1

. The open-weight AI model features a 1-million-token context window and represents a significant upgrade from the company's July release of Kimi K32

2

. With a score of 55 on Artificial Analysis Intelligence Index, Kimi K3.1 Pro ranks above Claude Fable 5.1 and GPT-6 Astra, which both scored 53, establishing it as the strongest model out of China4

4

.

Advanced Architecture Powers Superior Performance

The AI model comprises 896 individual neural networks with 2.8 trillion total parameters, though it only activates 104 billion parameters per token, representing approximately 1.8 percent activation ratio4

4

. This selective activation approach minimizes inference costs while maintaining high performance. Kimi K3.1 Pro utilizes Kimi Delta Attention (KDA), a custom implementation of Gated DeltaNet that reduces RAM consumption in the model's attention mechanism1

1

. The architecture employs NoPE (No Positional Embeddings) throughout the entire network, completely omitting traditional RoPE layers, which facilitates a more flexible context window, better generalization, and improved computational efficiency4

4

. The model also uses MXFP4 weights with MXFP8 activations through quantization-aware training for massive hardware efficiency4

4

.

Benchmark Results Demonstrate Competitive Edge

On Terminal-Bench 4.0, a coding benchmark that evaluates AI models on real-world terminal-based tasks, Kimi K3.1 Pro scored 65.7%, surpassing GPT-6 Astra's 58.2% and approaching Claude Opus 5.5's 66.4%2

2

. The model scored 61.2% on Kimi Code Bench 3.0 Long-horizon, an internal benchmark testing multi-step execution and end-to-end problem solving, compared to K3's 55.95%2

2

. In software engineering evaluations, Kimi K3.1 consistently generated clean, optimized code for complex applications, outperforming both GPT-6 Astra and Opus 5.5 in speed and accuracy during the Software Engineering Marathon3

3

. The model also ranks 8th on Code Arena, matching Claude Opus 54

4

.

Excellence in 3D Modeling and Financial Analysis

Kimi K3.1 Pro delivers particularly competitive output quality in 3D modeling, scoring 92.7% on BenchCAD, a 3D model coding benchmark, placing it within four percentage points of Astra and Claude Opus 5.51

1

. The model generates 3D models by writing code snippets that specify object geometry in numerical terms rather than drawing them1

1

. It can create both static 3D models and animate them, with internal tests demonstrating animated simulations of human motion and multiple 3D video games1

1

. The model integrates seamlessly with widely used platforms including Blender, Unity, and Unreal Engine3

3

. In financial analysis, Kimi K3.1 Pro outperformed both GPT-6 Astra and Claude Opus 5.5 on OfficeQA Pro v2, a test measuring LLMs' ability to perform complex financial analysis tasks1

1

. It demonstrated unparalleled precision in processing intricate datasets during Frontier Finance tests, automating workflows with remarkable accuracy3

3

.

Cost Efficiency Democratizes Access to Advanced AI

Moonshot AI has priced Kimi K3.1 Pro at $2 per million input tokens, $10 per million output tokens, and $0.20 per million cache hit tokens, representing a 33 percent reduction compared to Kimi K3 across all three rates4

4

. This pricing structure makes output costs approximately 80% less expensive than GPT-6 Astra, offering significant savings for businesses and developers3

3

. The affordability enables organizations of all sizes to use advanced AI capabilities without exceeding budgets, opening doors for smaller organizations, independent developers, and startups3

3

. A high-speed variant, Kimi K3.1 Pro Highspeed, delivers 2-3x the speed of K3.1 Pro for more responsive interactions while maintaining the same high-quality outputs2

2

.

Strategic Cybersecurity Trade-off and Future Plans

Kimi K3.1 Pro's performance on popular cybersecurity benchmarks significantly trails the scores of Astra and Opus 5.51

1

. Moonshot AI revealed that engineers deliberately identified and removed cyber-related data from the model's post-training dataset to address safety risks1

1

4

4

. The company plans to equip future models with more advanced vulnerability analysis features1

1

. Kimi K3.1 Pro is available via Moonshot AI's paid cloud services across Kimi.ai, Kimi mobile apps, Kimi Work, Kimi Code, and through the Kimi API2

2

. The full model weights release is scheduled for November 4, enabling developers and researchers to customize and optimize the model for specific requirements2

2

. This open-sourcing strategy positions Kimi K3.1 Pro as a noteworthy advancement in accessible AI development3

3

.