Moonshot AI has launched Kimi K3.1 Pro, a multimodal mixture-of-experts model with 2.8 trillion total parameters, 104 billion active per token, and a 1-million-token context window. The company says the model is a post-trained version of Kimi K3 with substantially more post-training compute, and it highlights gains in full-project coding, Blender and CAD work, financial modelling, and long-running agents. Standard and Highspeed versions are available now through Kimi's website and app, Kimi Work, Kimi Code, and the API. Moonshot says Highspeed generates output two to three times faster without a drop in intelligence, though its official account says it costs twice as much. The launch also includes Kimi Hosted Agents, a managed harness with a 24/7 execution environment for long-running and multi-agent work. Independent evaluators place the model near the top of the field but not at it. Artificial Analysis scored K3.1 Pro 55 on its Intelligence Index, behind Opus 5.5 at 58 and Sonnet 5.5 and Muse Wonder Full at 56, with its strongest gains in agentic tasks. Its 64 percent score on Artificial Analysis's Cyber Index is the highest result among publicly accessible models and second overall, behind the trusted-access GPT-6 Sol Daybreak Blue at 69 percent, although Moonshot says it deliberately reduced cyber-related training. On Arena's WebDev leaderboard the model ranks eighth, with 1,692 points, up nine places from Kimi K3. Commentators have been divided. Some praised the prospect of an open-weight model competitive with leading closed systems, while others questioned inference costs, weaker results on some interactive tasks, benchmark presentation, and allegations of undisclosed paid promotion. The weights are not yet public. Moonshot has promised an open-weight release on November 4, 2026, but the licence, redistribution terms, and commercial-use conditions have not been published, so calling the model fully open source would be premature. Deployment is also far from trivial: 2.8 trillion parameters implies roughly 5.6 TB of storage at 16-bit precision or about 1.4 TB at 4-bit before serving overhead, so the 104 billion active-parameter figure does not describe the hardware needed to run it. For Australian businesses and professionals, nothing changes immediately. The API is available now and can be tested on non-sensitive work, but any use involving personal information sent to a Chinese-hosted service should be reviewed against the Privacy Act's cross-border disclosure rules (APP 8) and the organisation's own data-handling policy. Self-hosting is not realistic for most organisations, and the Highspeed tier's higher price should be weighed against actual latency needs before adoption.