# apex-flash-1

apex-flash-1 is Cantina Security's first open-weights security model, developed in partnership with Yeta (@yetalabs on X). It is a reinforcement learning post-train of GLM-5.3-Flash for focused investigations: reading code, using tools, pursuing an exploit, and verifying its effect in a running…

- **Kind:** model (Hugging Face)
- **Hugging Face:** https://huggingface.co/cantina-security/apex-flash-1
- **Owner:** cantina-security (https://forefy.com/u/hf/cantina-security)
- **Parameters:** 321.3B
- **Context:** 1048576 tokens
- **License:** mit
- **Focus:** offensive
- **Architecture:** Glm5NextForConditionalGeneration
- **Pinned revision:** `28c647a6bd0444973a7ef3d940c24c67e64f52fe`
- **Page:** https://forefy.com/models/f6308f0c-2694-5a2a-9ce6-e04337dad13c
- **API (JSON):** https://forefy.com/api/asr/models/f6308f0c-2694-5a2a-9ce6-e04337dad13c

## What was done to it

- finetune (declared evidence, from zai-org/GLM-5.3-Flash)

## Lineage

zai-org/GLM-5.3-Flash → cantina-security/apex-flash-1

## Checks (repo at the pinned revision; not a behavior test)

- **Identity** [pass]: No lookalike name or conflicting provenance
- **Weights format** [pass]: safetensors only, nothing runs on load
- **Pickle files** [pass]: None.
- **Remote code** [pass]: None
- **Chat template** [warn]: Differs from zai-org/GLM-5.3-Flash's
- **Refusals modified** [pass]: No
- **License** [pass]: mit
- **Lineage** [pass]: Derives from zai-org/GLM-5.3-Flash
- **Model card** [warn]: 3 of 6, missing training data, evaluation results, intended use and limitations
- **Safetensors headers** [pass]: Every byte is indexed, no hidden data (8 of 62 files read)
- **Tensor layout** [warn]: 0 tensor(s) added, 37338 removed vs zai-org/GLM-5.3-Flash (partial read)

## Variants

- mixed safetensors (safetensors, 642.7 GB)

## Run it

- transformers: `AutoModelForCausalLM.from_pretrained("cantina-security/apex-flash-1", revision="28c647a6bd0444973a7ef3d940c24c67e64f52fe")`