Tech • AI • Robotics • Game
I tested Kimi 3.1 Pro: The end of OPUS 5.5? My test!
7/10TL;DR
Moonshot AI’s Kimi 3.1 Pro enters the top tier of multimodal models with strong coding, 3D and finance performance, but its pricing, speed and gameplay execution still trail the best OpenAI alternatives in some practical tests.
KEY POINTS
Positioning in the market
Kimi 3.1 Pro was presented as one of the strongest new general-purpose models, with benchmark performance near Claude Opus 3.5 in professional tasks and just behind leading coding systems. Its launch immediately places it among the more capable premium models rather than low-cost utility offerings.
Pricing remains premium
The model is priced at about $2 per million input tokens and $10 per million output tokens. That makes it cheaper than GPT Astra by roughly a factor of five, but still noticeably more expensive than lower-cost rivals such as GPT 6.1-class models, leaving it outside the most cost-efficient segment.
PDF understanding is still weak across the industry
One notable benchmark highlighted a score near 34% on PDF comprehension tasks. That implies high error rates when users feed raw PDFs directly into AI systems, reinforcing the need to convert documents into structured text or data before analysis.
3D and game-development training stand out
Kimi 3.1 Pro appears heavily optimized for tools and workflows related to Unreal Engine, Godot, Blender and CAD-like scene control. In testing, it generated 3D assets, configured software, and assembled a game prototype in roughly 25 minutes, suggesting unusually strong model training for graphics and interactive content pipelines.
Game prototype looked impressive but needed fixes
In a direct coding test, Kimi 3.1 Pro produced a visually polished Paris-themed game environment with strong textures, lighting changes, map elements and collectible objects. However, the first build had launch and movement issues, requiring extra iteration before the game became partially playable.
OpenAI performed better on playability
A comparable game built with ChatGPT 6.1 was reported to be fully playable in a single pass, with working movement, jumping, quest collection and smoother cinematic presentation. The comparison suggested that Kimi 3.1 Pro currently excels more in visual fidelity than in reliable one-shot gameplay logic.
Kimi showed strong texture and scene-detail choices
In side-by-side 3D city generation, Kimi 3.1 Pro delivered detailed trees, brickwork, windows and ground textures, making different but credible artistic choices from OpenAI outputs. Even where its navigation and automated camera behavior were weaker, its raw scene rendering showed clear progress over earlier Kimi 3 versions.
Web interface generation was a practical success
Beyond gaming, the model generated an Amazon-style ecommerce interface with animated elements, product listings, cart functionality and a more marketing-driven visual design. Tests indicated that adding and removing items worked, showing that Kimi 3.1 Pro can handle full UI assembly in near one-shot website prototyping.
Built-in scene editing could help developers
A notable feature is its ability to expose a scene tree and let users select objects visually, identify where they sit in the map, and request edits to precise zones. That combination of code, layout awareness and object-level interaction could be valuable for rapid iteration in development workflows.
Finance and spreadsheet use are another target area
The model was also described as highly competitive on finance benchmarks, with strong positioning for agent-based analysis, crypto-related workflows and spreadsheet tasks. Support for browser-based file handling, including Excel-type use cases, broadens its enterprise appeal beyond coding and graphics.
Speed and usage controls matter
In autonomous computer-use mode, the model could take over the interface and run tests, but response speed was slow at default settings. Switching to a high-speed mode significantly improved performance, though it likely increased consumption; one intensive session reportedly used about 16% of a $39 Pro subscription.
Data sovereignty remains a concern
Because Kimi 3.1 Pro is a Chinese model, data handling is a central issue for professional users. Those seeking U.S.-hosted routing can use intermediaries such as OpenRouter or other paid infrastructure providers, which may reduce direct exposure to China-based processing.
CONCLUSION
Kimi 3.1 Pro looks like a serious new contender for advanced coding, 3D production and finance tasks. Its strongest edge is creative and technical scene generation, while its biggest hurdles are premium cost, slower autonomous operation and less reliable one-shot execution than the best OpenAI systems.
Ask a question