Kimi K3.1 Pro Narrows the Gap With Top AI Models While Cutting Costs

Moonshot AI has released its Kimi K3.1 Pro reasoning model, which ranks fourth on the Artificial Analysis Intelligence Index with 55 points. This score is a gain of 11 points over its predecessor, Kimi K3, and places it just three points behind the leading proprietary models. The company has also released a faster variant, K3.1 Pro Highspeed, which runs two to three times quicker. Both models are available now through Kimi.ai, Kimi Work, Kimi Code, and the Kimi API.

Kimi K3.1 Pro is a post-trained version of Kimi K3 that has received significantly more post-training compute. The model's parameter count remains at 2.8 trillion total parameters with 104 billion active ones, and its context window spans one million tokens. According to Artificial Analysis, no model with a comparable parameter count achieves a higher score on the Intelligence Index.

The model's biggest improvements appear in agentic tasks, coding, and 3D modeling. On the GDPval-AA benchmark, K3.1 Pro reaches 1,846 Elo, an increase of 312 points over K3. This places it just behind Claude Opus 5.5, which scores 1,866. On Terminal-Bench 4.0, Moonshot reports a score of 65.7 percent for K3.1 Pro, compared to 66.4 percent for Claude Opus 5.5.

Hallucination rates have also dropped sharply. On Moonshot's internal Kimi Hallu Bench Hard, claim-level hallucinations fell from 23.4 percent with K3 to 8.3 percent with the new model. The company notes this is a stress test rather than a measure of general error rates. GPT-6 Astra scores 5.5 percent on the same benchmark.

For 3D modeling and coding, K3.1 Pro interacts directly with game engines such as Unity, Unreal Engine, and Godot. On the BenchCAD benchmark, it scores 92.7 percent with tools, close to Claude Opus 5.5's 96.2 percent. Without tools, K3.1 Pro reaches 72.8 percent, nearly matching Claude Opus 5.5's 73.0 percent. The company states the model is strongest at parametric modeling of mechanical parts, architectural models, and configurable products.

Alongside these capability gains, API prices for K3.1 Pro have dropped by 33 percent. Input costs are now $2 per million tokens, down from $3, while output costs are $10 per million tokens, down from $15. Each Intelligence Index task costs $1.98 to run, almost identical to the $2.00 cost for K3, as the model generates 14 percent more output tokens.

Moonshot also addressed cybersecurity concerns in its technical blog post, stating that vulnerability research capabilities grew as post-training scaled. For the open-weights release, the company removed cyber-related data from both training phases. On its internal xdev-bench-hard, which reproduces recently disclosed vulnerabilities in software like the Linux kernel and Node.js, the uncleaned version solves 90.5 percent of tasks. After data cleaning, K3.1 Pro solves 81.0 percent, while DeepSeek-V4.1-Flash reaches 85.7 percent.

Moonshot plans to release the model weights by November 4, 2026. If that happens, K3.1 Pro would become the highest-scoring open-weights model available, narrowing the gap between open and proprietary models from 12 points to just 3 points. The model is available for use now through Kimi.ai, Kimi Work, Kimi Code, and the Kimi API.

As of October 23, 2026, K3.1 Pro remains unavailable as open weights, and Moonshot has not announced a specific date beyond the November 4, 2026 target for their release.