Moonshot AI Launches Kimi K3.1 Pro With Free Access, Strong Coding Scores, and Reduced Cybersecurity Risk
Moonshot AI has introduced Kimi K3.1 Pro, the first model in its new K3.1 lineup, and it is already drawing attention for its combination of high-end coding performance, a massive context window, and an unusual focus on limiting cybersecurity misuse.
The model is now available through Kimi.ai, the Kimi apps for iOS and Android, the Kimi Work desktop app, the Kimi Code programming tool, and the Kimi API. Users can try Kimi K3.1 Pro for free, although Moonshot AI describes the free tier as offering “limited usage” without clearly stating the exact allowance. For users who want faster responses, Kimi K3.1 Pro Highspeed is available on paid plans and is said to run two to three times faster than the standard version.
Kimi K3.1 Pro is based on Kimi K3, which launched in July, but Moonshot AI says the newer version received significantly more post-training compute. The company describes it as a mixture-of-experts AI model with 2.8 trillion total parameters, of which 104 billion are active for each request. One of its standout features is a 1 million-token context window, allowing it to process very large documents, long codebases, extended conversations, and complex multi-step tasks.
For developers, the biggest highlight is coding performance. Moonshot AI claims Kimi K3.1 Pro achieved 65.7% on Terminal-Bench 4.0, a benchmark designed to test AI coding agents on terminal-based tasks. That puts it close to Claude Opus 5.5, which the company lists at 66.4%, and ahead of GPT-6 Astra at 58.2%. These figures come from Moonshot AI, so they should be treated as company-provided results rather than fully independent verification.
On the Artificial Analysis Intelligence Index, Kimi K3.1 Pro reportedly scores 55, compared with 58 for Claude Opus 5.5. While that still leaves it behind one of the strongest closed models, it suggests Moonshot AI is pushing Kimi into the upper tier of current AI coding systems.
Pricing is also part of the appeal. API access for Kimi K3.1 Pro is listed at $2 per million input tokens and $10 per million output tokens. Moonshot says this is about one-third cheaper than Kimi K3, making the new model potentially attractive for developers, startups, and companies looking for advanced AI coding capabilities without paying premium frontier-model prices.
What makes this release especially notable is Moonshot AI’s public discussion of cybersecurity risk. The company says that as it scaled post-training, Kimi K3.1 Pro became increasingly capable at finding and exploiting software vulnerabilities. According to Moonshot, this improvement came from cyber-related material in the general training data and from cybersecurity tasks included in reinforcement learning environments.
For the upcoming open-weight release, Moonshot says it identified and removed cyber-related data from both stages of training while preserving the general coding ability needed for software development. The goal, according to the company, is to avoid releasing an open model that pushes the absolute frontier of cyber capability.
Moonshot tested the impact of these changes using an internal benchmark called xdev-bench-hard. In this test, an AI agent attempts to exploit recently disclosed vulnerabilities in software such as the Linux kernel and Node.js. A checkpoint trained without the new restrictions reportedly solved 90.5% of the tasks, while the final Kimi K3.1 Pro solved 81%.
That reduction shows the safety measures had an effect, but it also shows the model still retains substantial vulnerability exploitation ability. For comparison, the original Kimi K3 reportedly scored 33.3% on the same internal test. Moonshot says Kimi K3.1 Pro remains behind closed frontier models that have already been publicly available for at least three months, though these cybersecurity benchmark results have not been independently confirmed.
The company plans to release the full Kimi K3.1 Pro model weights by November 4. This is a major step because once open weights are published, they cannot realistically be taken back. Other developers can modify the model, remove built-in safeguards, or deploy it in ways the original creator may not intend.
That is why Moonshot’s decision to reduce attack-related training data is important. Safety filters can often be removed after release, but capabilities that were never deeply trained into the model are harder to restore. In other words, Moonshot appears to be trying to lower the risk at the model capability level, not just through surface-level restrictions.
Even so, Kimi K3.1 Pro is not a model most people will run locally. With 2.8 trillion total parameters, it is far beyond the reach of a typical consumer PC or laptop. Most users will access it through Moonshot’s cloud services, including the web app, mobile apps, desktop app, coding tool, or API.
That also raises privacy considerations. Moonshot mentions a zero data retention option for API customers, but everyday users trying the free web or app versions should still think carefully before submitting sensitive business data, private code, personal documents, or confidential files to any cloud AI service.
Kimi K3.1 Pro arrives at a time when AI labs are racing to build better coding agents, longer-context models, and more capable open systems. Moonshot AI’s new release stands out because it combines strong programming performance, competitive pricing, free access, and a more transparent discussion of cybersecurity concerns than many AI launches.
For users, the model could be useful for coding assistance, debugging, software planning, document analysis, research, and long-context productivity tasks. For the AI industry, it is another sign that open and semi-open models are quickly closing the gap with leading closed systems, especially in programming and agentic workflows.
The central question is whether Moonshot AI can balance accessibility with safety. Kimi K3.1 Pro appears powerful enough to attract developers and AI enthusiasts, but its cybersecurity capabilities will likely remain under close scrutiny as the open-weight release approaches.
