AI Ranking Shift: Moonshot's New Model Cuts Lead

A major shift in the artificial intelligence hierarchy occurred today as Moonshot AI unveiled Kimi K3.1 Pro. The new reasoning model immediately secured the fourth position on the Artificial Analysis Intelligence Index. This leap in status places the Chinese firm significantly closer to the leading proprietary systems. The update marks a pivotal moment for open-weight competitors in a market long dominated by closed-source giants.

The performance data reveals a dramatic uptick in capability. K3.1 Pro achieved a total score of 55 points. This figure represents an 11 point increase over its direct predecessor, Kimi K3. The gap between this new entrant and the top of the leaderboard has narrowed to just three points. Only three models currently hold higher scores on the index: Anthropic’s Claude Opus 5.5 with 58 points, Claude Sonnet 5.5 with 56 points, and Muse Wonder Full also at 56 points. This clustering at the top suggests a tightening race among the highest performing systems available.

Technical details indicate that the improvement stems from intensified post-training. Moonshot applied substantially more compute during this phase compared to the original K3 release. The company has scheduled the public release of the model weights for November 4, 2026. If this timeline holds, K3.1 Pro will become the highest scoring open-weights model on the market. Such a release would shrink the disparity between open and proprietary models from 12 points to three, provided no other major updates occur before that date. Users can access both the standard version and the K3.1 Pro Highspeed variant through various platforms including Kimi.ai, Kimi Work, Kimi Code, and the Kimi API. The Highspeed edition is designed for speed, running two to three times faster than the main model.

Performance gains were most visible in agentic tasks and complex reasoning. On the GDPval-AA benchmark, the new model reached an Elo rating of 1,846. This is a 312 point jump from the previous iteration. It trails only Claude Opus 5.5 at 1,866 and Muse Wonder Full at 1,857. Coding capabilities also saw a significant boost. On Terminal-Bench 4.0, Moonshot reports a score of 65.7 percent. This places it ahead of GPT-6 Astra, which scored 58.2 percent, and just below Claude Opus 5.5 at 66.4 percent. Notably, the older K3 model managed only 13 percent on this specific test, highlighting the magnitude of the recent upgrade.

Reliability has also improved according to internal testing. Hallucination rates dropped sharply in the company’s Kimi Hallu Bench Hard evaluation. Claim-level hallucinations fell from 23.4 percent in the previous model to 8.3 percent in the new version. Moonshot describes this benchmark as a stress test rather than a general measure of error. GPT-6 Astra remains the leader on this metric with a rate of 5.5 percent. Financial analysis capabilities also advanced, with gains of 7.2 points in Financial Modeling and 11.6 points in Financial Research. The company claims the model now tops all tested systems on OfficeQA Pro v2.

Specialized applications in 3D modeling and coding received further attention. The model integrates with major game engines like Unity, Unreal Engine, and Godot. It can also write and execute scripts for Blender and CAD software without requiring MCP connectors. On the BenchCAD benchmark, the system scored 92.7 percent when using tools. This compares to 96.2 percent for Claude Opus 5.5 and 95.9 percent for GPT-6 Astra. Without external tools, the score was 72.8 percent, nearly matching the 73.0 percent achieved by Claude Opus 5.5. A new 3D Make feature on Kimi.com now allows users to preview, edit, and export 3D models directly.

The underlying architecture remains consistent with the K3 base. It features 2.8 trillion total parameters, with 104 billion active. The context window supports one million tokens and handles both text and image inputs. Artificial Analysis places it at the top of the Pareto frontier, indicating no model of similar size scores higher. Pricing has become more competitive as well. API costs dropped by 33 percent, setting input at $2 per million tokens, output at $10, and cache hits at $0.20. Despite generating 14 percent more output tokens, the cost per Intelligence Index task remains stable at $1.98, nearly identical to the previous $2.00.

Security considerations played a key role in the development process. Moonshot noted that vulnerability research capabilities improved as post-training expanded. Cyber-related content was part of the mid-training corpus and included in the reinforcement learning mix. For the open-weights release, the team removed this data from both training stages to balance offense and defense. In internal testing on xdev-bench-hard, which replicates recent vulnerabilities in software like the Linux kernel and Node.js, the uncleaned model solved 90.5 percent of tasks. After data cleaning, this figure dropped to 81.0 percent. DeepSeek-V4.1-Flash scored 85.7 percent, while the older Kimi K3 managed only 33.3 percent. The company aims to gradually raise the capabilities of open models to support defensive auditing, even as it avoids pushing the absolute limits of offensive cyber capabilities.

The next few weeks will be critical as the community prepares for the open-weights release. Analysts will likely scrutinize the real-world performance of the model against its benchmark claims. The narrowing gap between open and proprietary systems may accelerate adoption in enterprise settings. For now, the focus remains on November 4, when the weights are expected to become publicly available.