Kimi K3.1 Pro Narrows Gap With Top AI Rivals

55 points on the Artificial Analysis Intelligence Index marks the new benchmark for Kimi K3.1 Pro, a reasoning model released by Chinese AI firm Moonshot AI. This score places the system in fourth position globally, trailing only three proprietary models: Claude Opus 5.5 at 58 points, and Claude Sonnet 5.5 and Muse Wonder Full, both at 56. The 11-point leap from its predecessor, Kimi K3, signals a significant shift in the competitive landscape for large language models.

If Moonshot adheres to its plan to release the model weights by November 4, 2026, K3.1 Pro will become the highest-scoring open-weights model available. Such a move would compress the performance gap between open and proprietary systems from 12 points to just three, assuming no other major releases occur in the interim. A faster variant, K3.1 Pro Highspeed, which runs two to three times quicker, is also currently accessible via Kimi.ai, Kimi Work, Kimi Code, and the company's API.

The most substantial improvements appear in agentic tasks. On the GDPval-AA benchmark, the model achieves 1,846 Elo, a 312-point increase over K3, placing it just behind Claude Opus 5.5 and Muse Wonder Full. In coding evaluations, K3.1 Pro records 65.7 percent on Terminal-Bench 4.0, surpassing GPT-6 Astra while remaining close to Claude Opus 5.5. This represents a dramatic recovery from K3’s 13 percent score on the same test. Hallucination rates have also declined sharply; on Moonshot’s internal stress test, claim-level errors dropped from 23.4 percent to 8.3 percent, though GPT-6 Astra maintains a lower rate of 5.5 percent.

Capabilities in 3D modeling and coding have expanded significantly. The model now interacts directly with game engines such as Unity, Unreal Engine, and Godot for both minor fixes and full project builds. On the BenchCAD benchmark, it scores 92.7 percent with tool assistance and 72.8 percent without, nearly matching the top proprietary competitors. Moonshot notes its strength in parametric modeling for mechanical and architectural designs, and has launched a "3D Make" feature on Kimi.com that allows users to preview, edit, and export 3D assets.

Despite the performance gains, the architectural footprint remains unchanged. K3.1 Pro retains the 2.8 trillion total parameters and 104 billion active parameters of its predecessor, with a one-million-token context window. However, API prices have dropped by 33 percent. Input costs are now $2 per million tokens, down from $3, while output costs stand at $10, reduced from $15. Cache hit costs have also fallen to $0.20 from $0.30. Consequently, the cost per Intelligence Index task is $1.98, virtually identical to the previous version.

Moonshot has also addressed cybersecurity concerns in its technical documentation. The company observed that vulnerability research capabilities grew as post-training scaled, prompting them to strip cyber-related data from the training phases for the open-weights release. On an internal benchmark reproducing recent software vulnerabilities, the cleaned model solves 81.0 percent of tasks, down from 90.5 percent for the uncleaned version. While DeepSeek-V4.1-Flash scores 85.7 percent on the same test, Moonshot aims to gradually expand open-model capabilities for defensive purposes like auditing legacy systems.

The release underscores a trend toward balancing efficiency with performance in the AI sector. By maintaining the same parameter count while lowering costs and improving specific task outcomes, Moonshot positions itself as a key player in the open-weights market. The upcoming November release will be a critical test of whether open models can sustainably close the gap with leading proprietary systems.