What happened
Moonshot AI released Kimi K3.1 Pro, a mixture-of-experts model with 2.8 trillion total parameters and 104 billion active parameters, featuring native multimodal support and a 1-million-token . The model is currently available via the Kimi API, Kimi.ai, Kimi Work, and Kimi Code, with open weights scheduled for release by November 4, 2026. A faster variant, Kimi K3.1 Pro Highspeed, is also available, running 2-3 times quicker than the standard version. Moonshot reports that K3.1 Pro scored 65.7% on Terminal-Bench 4.0, surpassing GPT-6 Astra (58.2%) but trailing Claude Opus 5.5 (66.4%). The model is priced at two-thirds of the previous K3 generation, with standard input/output costs roughly half of Claude Opus 5.5 and a fraction of GPT-6 Astra.
Moonshot AI has released Kimi K3.1 Pro, an upgraded version of its open-weight model series. The model is a mixture-of-experts (MoE) architecture with 2.8 trillion total parameters and 104 billion active parameters. It supports native multimodal inputs and offers a of up to 1 million tokens. A faster variant, Kimi K3.1 Pro Highspeed, is also available, which Moonshot states runs 2-3 times quicker than the standard version.
Both versions are currently live on Kimi.ai, Kimi Work, Kimi Code, and the Kimi API. Moonshot has promised to release the open weights for Kimi K3.1 Pro by November 4, 2026. The company positions the model for full-project creative and technical workflows, including coding in game engines like Unreal, Unity, and Godot, as well as parametric 3D modeling in Blender and CAD software.
According to OfficeChai, Moonshot's internal and third-party benchmark results show Kimi K3.1 Pro scoring 65.7% on Terminal-Bench 4.0, which tests coding agents on realistic terminal tasks. This score places it ahead of OpenAI's GPT-6 Astra (58.2%) but slightly behind Anthropic's Claude Opus 5.5 (66.4%). On BenchCAD, which evaluates 3D reconstruction from multi-view renderings, K3.1 Pro scored 92.7% with tool use, close to GPT-6 Astra's 95.9% and Claude Opus 5.5's 96.2%.
The model is priced at two-thirds of the cost of the original K3 model. Standard K3.1 Pro input and output pricing is reported to be roughly half of Claude Opus 5.5's $4/$20 per million tokens and a fraction of GPT-6 Astra's $10/$50 per million tokens. The Highspeed variant is priced at $4 input and $20 output per million tokens, matching Opus 5.5's price point while offering higher speed.
Source details: officechai.com ↗
Why it matters
This release intensifies competition in the frontier AI market by offering near-flagship performance at significantly lower costs, challenging the pricing and performance dominance of US-based labs. The commitment to open weights by November 4, 2026, is particularly significant as it allows independent verification of the model's capabilities and facilitates broader adoption in research and enterprise settings. The reported improvements in financial modeling reliability and full-project coding workflows suggest a shift toward more practical, specialized AI applications beyond general chatbot interactions.
The release of Kimi K3.1 Pro represents a significant competitive move by a Chinese AI startup, offering performance that rivals top proprietary models from Anthropic and OpenAI at a lower cost. This challenges the traditional pricing structures of Western frontier labs and may accelerate the adoption of open-weight models in enterprise environments.
The specific focus on financial modeling and creative workflows, with reported 65% reduction in hallucinations for financial tasks compared to the previous generation, suggests a maturation of AI capabilities in specialized, high-stakes industries. This could lead to broader integration of AI in finance and design sectors.
The promise of open weights by November 4, 2026, is crucial for the AI community as it allows for independent verification of the model's capabilities. This transparency can help build trust and facilitate further research and development based on the model's architecture.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
crm_get_transaction(id='4092').Which component of an AI application is the machine-learning model itself?
What to watch next
Monitor the actual release of open weights on November 4, 2026, to verify if independent benchmarks align with Moonshot's claims. Watch for enterprise adoption trends, particularly in finance and creative industries, to see if the lower pricing and specialized capabilities drive significant market share shifts. Additionally, observe how Anthropic and OpenAI respond to this price-performance pressure in their subsequent model updates or pricing adjustments.
The actual release of open weights on November 4, 2026, will be a key milestone. Independent benchmarks will be necessary to confirm whether the model's performance matches Moonshot's claims, particularly in areas like coding and financial modeling.
Market reaction from enterprises in finance and creative industries will indicate whether the lower pricing and specialized capabilities are sufficient to drive significant adoption. Monitoring API usage trends and enterprise partnerships will provide insights into real-world impact.
Responses from Anthropic and OpenAI, such as new model releases or pricing adjustments, will be important to gauge the competitive dynamics in the frontier AI market. The ability of these companies to maintain their performance and pricing advantages will be a critical factor in the ongoing competition.