Supported AI Models
PacketSafari tests alternative models for both compatibility and investigation quality. A model must produce dependable, packet-grounded results before it is enabled for production Agent use.
| Model | Current guidance |
|---|---|
| Kimi K2 | Supported and recommended · Fast. The recommended alternative-model starting point for qualified customer-operated deployments. |
| Kimi K2.6 | Supported · Fast. |
| DeepSeek V4 Pro | Supported · Fast. |
| DeepSeek V4 Flash 0731 | Supported · Fast. |
| GLM 5.2 | Supported after deployment qualification. Available as an alternative model for qualified deployments. |
| Kimi K3 | Promising candidate. Available for assessment in high-capacity private-AI deployments. |
| Qwen3.7 Flash | Not supported. |
| Qwen3.6 27B | Not supported. |
| Qwen3 Coder 30B | Not supported. |
| Qwen3 235B | Not supported. |
| Qwen3 Coder (large) | Not supported. |
| GLM 4.7 Flash | Not supported. |
| DeepSeek V4 Flash (earlier tested route) | Not supported. |
| MiniMax M3 | Not supported. |
Fast identifies a tested route that answered the evaluated focused case correctly within the 60-second target. It is not a universal latency guarantee.
For a current alternative-model pilot, start with Kimi K2, Kimi K2.6, DeepSeek V4 Pro, or DeepSeek V4 Flash 0731. GLM 5.2 is another qualified deployment option. Models marked not supported failed qualification or PacketSafari investigation-quality acceptance and remain unavailable for production Agent stages.
For Kimi K3 capacity planning, 8× B300 with 2,304 GB aggregate GPU memory is
a reference profile for deployment assessment. Final model and workload fit is
confirmed during solution design.
Each customer-operated deployment is qualified for its selected model, infrastructure, and workload before production use. Additional model families can be evaluated during deployment validation.