Skip to content

Model Performance

Lynx AI Agent is multi-model by design. This flexibility allows you to try and switch between AI models and find the one that best suits your needs.

Choosing the right model is not always straightforward. Models can vary in feel, intelligence, alignment, reasoning abilities, and more.
That's why you, as a Lynx AI power user, are encouraged to try different AI models included in your subscription for various Splunk tasks and use cases.

AI Model Comparison

Because researching the characteristics of different AI models is extensive and time-consuming, we have compiled an opinionated list of officially supported AI models for Lynx AI Agent.
You're welcome to use it as a starting point when choosing an AI model for your use case.

The models are rated on several different categories:

  • Splunk Knowledge: How well the model understands Splunk concepts, SPL, and common workflows.
  • Context Retrieval: Reliability when invoking tools and acting on tool outputs.
  • Dashboard Generation: Quality of generated dashboards, panels, and visualizations.
  • Data Intelligence: The ability of the model to interpret results, spot patterns, and suggest next steps.

Note

These ratings are opinionated and can change over time, as the state-of-the-art moves forward and more AI models are introduced across different fields of expertise.
Feel free to check back from time to time, or to conduct your own experiments with the different models.

Info

Some models support reasoning 💡 (thinking) mode, non-reasoning mode, or both.
Reasoning models tend to perform better on complex tasks, but may be slower and use more tokens.

Proprietary Models

These models are proprietary and are only available on the cloud version of Lynx AI Agent.
They don't require Splunk Cloud; they only require having an internet connection available.

Model Splunk Knowledge Context Retrieval Dashboard Generation Data Intelligence
Claude Sonnet 4.6 🟢 🟢 🟢 🟢
Gemini 3.1 Pro 🟡 🟡 🟢 🟡

Open-Weight Models

These models are open-weight, meaning they are available for the public to use and self-host. They are available for Lynx AI Agent in on-premises deployments, and some may also be available in the cloud offering.

Note

The open-weight models feature another column called Resource Requirements. It illustrates how easy it is to self-host the model on enterprise-grade GPUs (relatively).

Model Splunk Knowledge Context Retrieval Dashboard Generation Data Intelligence Resource Requirements
Kimi K2.7 Code 🟢 🟢 🟢 🟡 🟡
GLM 5.2 🟡 🟢 🟡 🟢 🔴
GLM 4.7 🟡 🟡 🟡 🟢 🟡
GLM 4.7 Flash 🟡 🟡 🟡 🔴 🟢
Qwen 3.5 397B 🟡 🟡 🟡 🟢 🟡
MiniMax M3 🟡 🟢 🟡 🟢 🟡
MiniMax M2.7 🟡 🟢 🟡 🟡 🟢
Gemma 4 31B 🟡 🟡 🟡 🔴 🟢
Gemma 4 26B 🟡 🟡 🔴 🔴 🟢

Models Under Evaluation

We're still evaluating the following models. We'll categorize them as soon as we have enough data to make a recommendation.

Model Splunk Knowledge Context Retrieval Dashboard Generation Data Intelligence Resource Requirements
Kimi K3 ⌛ ⌛ ⌛ ⌛ 🔴
Trinity Large Thinking ⌛ ⌛ ⌛ ⌛ 🟡

Unsupported Models

These models have been thoroughly tested, and are either deemed unsatisfactory for use in Lynx AI Agent, or outdated enough to be replaced by newer models that are both better and cheaper.

Model Splunk Knowledge Context Retrieval Dashboard Generation Data Intelligence Resource Requirements
GPT-OSS 120B 🟡 🔴 🔴 🟡 🟢
GPT-OSS 20B 🟡 🔴 🔴 🔴 🥔
Llama 4 Maverick 🔴 🔴 🔴 🔴 🟡
Llama 4 Scout 🔴 🔴 🔴 🔴 🟢
Llama 3.3 70B 🟡 🔴 🔴 🔴 🟢