Moonshot AI Launches Kimi K3.1 Pro to Challenge Top Rivals
The new reasoning model improves scores significantly and lowers prices, narrowing the performance gap between open and proprietary AI systems.
By Isabella Morales ·
Moonshot AI has released Kimi K3.1 Pro, a new reasoning model that now ranks fourth on the Artificial Analysis Intelligence Index with a score of 55. This places the model just three points behind the leading proprietary systems, marking a significant shift in the competitive landscape for artificial intelligence tools.
The new model represents an 11 point improvement over its predecessor, Kimi K3. According to Artificial Analysis, only three systems currently score higher than K3.1 Pro. These are Anthropic's Claude Opus 5.5 with 58 points, Claude Sonnet 5.5 with 56 points, and Muse Wonder Full with 56 points. Moonshot plans to publish the model weights by November 4, 2026. If this schedule holds, K3.1 Pro would become the highest scoring open weights model available. This move would reduce the performance gap between open and proprietary models from 12 points to just 3, assuming no other changes occur in the meantime.
Alongside the main release, the company is offering a variant called K3.1 Pro Highspeed. Moonshot states that this version runs two to three times faster than the standard model. Both versions are accessible through Kimi.ai, Kimi Work, Kimi Code, and the Kimi API. The architecture remains unchanged from the previous K3 version, featuring 2.8 trillion total parameters with 104 billion active parameters. It supports a one million token context window and processes both text and images.
The most significant performance gains appear in agentic tasks. On the GDPval-AA benchmark, K3.1 Pro reached an Elo score of 1,846. This is 312 points above the K3 score and sits just behind Claude Opus 5.5 at 1,866 and Muse Wonder Full at 1,857. In Terminal-Bench 4.0 testing, Moonshot reports a score of 65.7 percent. This result exceeds GPT-6 Astra at 58.2 percent and trails only Claude Opus 5.5 at 66.4 percent. The previous K3 model managed only 13 percent on this specific test.
Reliability has also improved, with a sharp drop in hallucinations reported by the company. On the internal Kimi Hallu Bench Hard, claim level hallucinations fell from 23.4 percent with K3 to 8.3 percent. Moonshot describes this benchmark as a stress test rather than a general error rate measure. GPT-6 Astra still leads this specific metric with a rate of 5.5 percent. Financial performance also saw gains, with the model gaining 7.2 points on Financial Modeling and 11.6 points on Financial Research. The company states it tops all tested models on OfficeQA Pro v2.
The model demonstrates strong capabilities in 3D modeling and coding. K3.1 Pro works with game engines such as Unity, Unreal Engine, and Godot. It can write and run scripts for Blender and CAD software without needing MCP connectors. On the BenchCAD benchmark, the model scores 92.7 percent with tools, compared to 96.2 percent for Claude Opus 5.5 and 95.9 percent for GPT-6 Astra. Without tools, it reaches 72.8 percent, which is nearly equal to 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.
Cost efficiency is a key focus for working developers and businesses. API prices have dropped by 33 percent, with input costs at $2 per million tokens and output at $10. Cache hits cost $0.20 per million tokens. Although the model generates 14 percent more output tokens, the total cost to run each Intelligence Index task is $1.98. This is nearly identical to the $2.00 cost for the K3 model, keeping the financial burden stable for users.
Moonshot also addressed cybersecurity in its technical documentation. The company notes that vulnerability research capabilities grew as post training scaled up. For the open weights release, cyber related data was removed from both training phases to manage capabilities. On the internal xdev-bench-hard test, the uncleaned model solved 90.5 percent of tasks, while the cleaned version solved 81.0 percent. DeepSeek-V4.1-Flash scored 85.7 percent, and the older Kimi K3 scored 33.3 percent.
The company argues that open weights models should not push the absolute frontier of offensive cyber capabilities, but should gradually improve to support defensive work like auditing legacy systems. Anthropic previously demonstrated that even models scoring 45 on the Artificial Analysis index, such as GLM-5.3, can write exploits at levels close to Claude Mythos. This context highlights the importance of balancing capability with safety in the open source ecosystem.
Source: The Decoder
