Herald of AI AIGenerated Thursday, November 12, 2026 EN · IT · ES
Herald of AI

AIGenerated AI news for everyone, written by AI and explained simply. Models, technology, culture and the people behind it.

New models

Kimi's new AI model approaches top US competitors in 3D design and matches them in coding

Kimi K3.1 Pro lets you build full games, 3D models, and financial analyses — now available to use today.

AIGenerated · 4 min read · Written by AI from the sources listed below
Close-up of blue 3D printed figurines on a printer bed, showcasing modern craftsmanship.
Close-up of blue 3D printed figurines on a printer bed, showcasing modern craftsmanship. File picture · Photo: Erik Mclean / Pexels

What you could do with it

  • A game developer could build complete games end-to-end in engines like Unity, Unreal Engine, or Godot by combining code generation with web search and 3D asset creation 1.
  • A designer could produce parametric 3D models — such as mechanical parts, architectural layouts, or custom furniture — and run Blender or CAD scripts directly without extra connectors 1.
  • A researcher or analyst could run end-to-end financial workflows, pulling data across real documents, verifying facts, and performing multi-step numerical reasoning 1.

Measuring up against the best

Kimi K3.1 Pro is a post-trained version of the earlier Kimi K3 model, meaning the team spent far more computing time helping it learn how to handle complex tasks. The company built a special training system so that the reinforcement learning process works smoothly across hundreds of different areas at once, delivering steady improvement rather than unpredictable jumps.

On coding benchmarks, K3.1 Pro scores 65.7% on Terminal-Bench 4.0, which tests how well AI agents solve real terminal-based tasks under tight constraints. That places it ahead of OpenAI's GPT-6 Astra at 58.2% and just below Anthropic's Claude Opus 5.5 at 66.4% 1. On Kimi's own internal coding benchmark, it scored 61.2%, up from the previous K3 model's 55.95% .

Building games and 3D worlds

Kimi K3.1 Pro can work across game development engines such as Unity, Unreal Engine, and Godot, combining code generation with web search and 3D asset creation to build complete experiences 1. It is also outstanding at parametric modeling — which means producing mechanical components, architectural layouts, or bespoke furniture — and can author and execute scripts for programs such as Blender and CAD software without requiring additional connectors .

In one demo, the model built a realistic human walking simulator on top of an open-source anatomical model containing 296 skeletal components and 402 muscle parts. Over 167 missions and 420 reviews, it diagnosed problems, corrected 36 calculation errors in shoulder-muscle mechanics, and got the simulated figure to walk continuously for 11.2 seconds 1.

The company also launched a new 3D modeling tool called 3D Make on kimi.ai, where users can preview, edit, and download models ready for 3D printing or share them as videos 1.

Financial research with fewer mistakes

Kimi K3.1 Pro also tackles end-to-end financial workflows. It achieves a score of 64.4 on OfficeQA Pro v2, an evaluation that examines how well the model can handle complicated fiscal data analysis across genuine financial documents using retrieval, cross-document verification, and multi-step numerical reasoning 1. The team reports big improvements over the earlier K3 model on financial research, financial modeling, and simulated business management tasks .

To reduce false information, K3.1 Pro's hallucination rate on challenging prompts dropped to 8.3% from 23.4% in the previous version 1.

What's available and when

Both K3.1 Pro and K3.1 Pro Highspeed are live today on Kimi.ai, Kimi Work, Kimi Code, and through the Kimi API 1. The full open-weight model files will be released by November 4 .

How good is it

On Terminal-Bench 4.0, which tests how well AI agents solve real terminal-based tasks under tight constraints, K3.1 Pro scored 65.7%, ahead of OpenAI's GPT-6 Astra at 58.2% and just below Anthropic's Claude Opus 5.5 at 66.4%, according to the company 1. On its own Kimi Code Bench, K3.1 Pro scored 61.2%, up from the previous K3 model's 55.95% . On the BenchCAD 3D reconstruction benchmark, the model posted a score of 92.7%, nearly matching Claude Opus 5.5 at 96.2% and GPT-6 Astra at 95.9%. For OfficeQA Pro v2, an evaluation that measures how well systems can analyze complex fiscal data using actual financial documents, it reached 64.4 — the best result available, according to the company .

Chart from the announcement: The chart compares AI model performance across benchmark categories using horizontal bars; no indication is given whether longer or shorter bars denote better performance

Picture: kimi.ai

Chart from the announcement: The BenchCAD chart compares model scores on 3D reconstruction tasks, with longer blue bars indicating higher performance across categories like Code, CAD, Design, and Web

Picture: kimi.ai

Chart from the announcement: The Finance chart compares multiple AI models on financial tasks using horizontal bars, where a longer bar means better performance

Picture: kimi.ai

Key facts

  • Kimi K3.1 Pro has a one-million-token context window and matches top US models on coding benchmarks.
  • K3.1 Pro Highspeed runs two to three times faster than the standard version.
  • Full open-weight model files will be available by November 4.
  • The model can build games, create 3D assets, and analyze financial documents.
  • Hallucination rate on hard prompts dropped from 23.4% to 8.3% compared to the previous K3 model; the share of responses containing at least one factual error fell to 42.4% of the predecessor's rate.

Sources

  1. Kimi K3.1 Pro Tech Blog: Advancing Open Intelligence

Keep reading