Smart Spectrum AI Releases Open Source GLM Model Family: MIT License, Z.ai Platform, and High-Speed Inference Service Unveiled

In a noteworthy move in China's large modeling space, Wiseparation AI recently released a new series of GLM large language models, covering 32B and 9B parameter scales, and open sourced them under the fully open MIT license. The new model family includes base, inference, and an innovative "contemplative" model, all of which have been made available through the newly enabled Z.ai The platform is open for free experience and is synchronized with the launch of its MaaS service platform bigmodel.cnThe

The move shows that Smart Spectrum AI is trying to expand its reach and user base in the fiercely competitive AI market through a combination of open source and high-performance commercialized services.

Open source modeling matrix: covering the basics to cutting-edge exploration

The core of this open source project is the 9B and 32B parameter scale models, all of which follow the MIT license agreement. This means that developers and enterprises can use these models for commercial purposes and distribute them freely for free, which greatly reduces the threshold of use and is expected to promote wider applications and community innovation.

Base model: GLM-4-32B-0414

With 32 billion parameters GLM-4-32B-0414 base model, which is claimed to compete in performance with mainstream domestic and international models with much larger parameter counts. The model was pre-trained using 15T of high-quality data with special inclusion of inference-based synthetic data. The post-training phase focuses on strengthening the capabilities of instruction following, code generation, and function calling through techniques such as human preference alignment, rejection sampling, and reinforcement learning.

Smart Spectrum AI emphasizes that the model excels in tasks such as engineering code generation, Artifacts generation (an interactive way of generating results), function calling, and report writing, with some benchmarking metrics approaching or even surpassing those of such GPT-4o cap (a poem) DeepSeek-V3-0324(671B parameters) and other models.

The model improves code processing capabilities, enabling the generation of single-file code with more complex structures.Z.ai The platform's dialog mode has a built-in preview feature that allows you to view the generated HTML and SVG results directly, making it easy for users to evaluate and iterate.

Example Prompt: Design a UI for a mobile machine learning platform that includes training tasks, storage management, and personal statistics. The personal statistics UI should show a graph of the user's resource usage over time. Use Tailwind CSS to beautify the page by tiling these 3 mobile UIs into a single HTML page.

Example Prompt: Design a drawing board that supports drawing custom functions, adding and removing custom functions, and assigning colors to functions.

Example Prompt: Demonstrate an LLM training flow with svg.

Reasoning Model: GLM-Z1-32B-0414

on the basis of GLM-4-32B-0414(math.) genusGLM-Z1-32B-0414 The inference model is deeply optimized for mathematical, code and logic tasks through cold-start and extended reinforcement learning strategies. Official data shows that its mathematical and complex problem solving abilities have been significantly enhanced, and the performance of some tasks can be compared with that of the 671B parameters. DeepSeek-R1 Comparable. Performance in benchmarks such as AIME 24/25, LiveCodeBench, GPQA, etc. confirms its potential for mathematical reasoning.

Lightweight Options: GLM-Z1-9B-0414

Notably, Smart Spectrum AI has also introduced a small-sized model with 9B parameters GLM-Z1-9B-0414. Despite the small number of parameters, it is claimed to still perform well on mathematical reasoning and general-purpose tasks, with performance at the top of open source models of the same size. This provides an option for scenarios that are resource constrained or require lightweight deployment with both efficiency and effectiveness.

Frontier Quest: GLM-Z1-Rumination-32B-0414 Contemplative Modeling

GLM-Z1-Rumination-32B-0414 It is positioned as an exploration of the future shape of AGI. Unlike traditional reasoning models, the "contemplative" model is designed to solve complex open problems through more steps of deep thinking. Its core feature is the ability to integrate search tools to process information during the thinking process, and to use rule-based rewards to guide reinforcement learning. The model supports a "ask-search-analyze-complete" research loop and is designed to enhance research-based writing and complex retrieval tasks. This represents the continued exploration of Wisdom Spectrum AI's model architecture and capability boundaries.

MaaS platform goes live in tandem: compelling performance and pricing strategy

In addition to the model open source, the base and inference models have been integrated into the Smart Spectrum MaaS open platform (bigmodel.cn), which provides API services.

The base model provides GLM-4-Air-250414 cap (a poem) GLM-4-Flash-250414 Two versions, the latter completely free, show an intention to entice users to try it out.

The inference model is then divided into three versions for different needs:

• GLM-Z1-AirX (Extreme Edition): claimed to be one of the fastest commercial inference models in China, with an actual inference speed of up to 200 Tokens/second. This speed has a significant advantage in scenarios that require real-time interaction or processing a large number of requests.
• GLM-Z1-Air (Cost Effective Edition): Competitively priced at just DeepSeek-R1 1/30th of a percent, targeting cost-sensitive, high-frequency invocation scenarios.
• GLM-Z1-Flash (free version): further lowering the barrier to use and driving wider developer adoption.

This tiered pricing strategy, combined with high-performance paid options and free introductory versions, is a common but effective means for large model providers to compete for market share.

Corresponding model price list:

New Interactive Portal: Z.ai Platform Launched

Smart Spectrum AI has launched a new domain name Z.aithat serves as the entry point for the interactive experience of its latest model. Select .ai The domain name itself has a strong brand signal, is concise and direct to the core business.

at the present time Z.ai The platform has three open source models online for users to experience:

• GLM-4-32B (base model): Experience code generation and Artifacts features.
• Z1-32B (Inference Model): Get a direct feel for inference speeds of up to 200 Tokens/second.
• Z1-Rumination-32B (contemplative modeling): exploring its deep research and complex task processing capabilities.

Overall, through this release, Smart Spectrum AI has demonstrated its comprehensive layout in model development, open source strategy and commercialization. In the increasingly crowded AI big model track, the combination of strong technical strength, flexible open source protocols and competitive commercialization solutions is the key to consolidating its market position and attracting developers and enterprise users.