Llama 4 Behemoth
Meta·
LLMsopen-weightlocalFrontier
Overview
2T MoE (288B active) teacher model. Announced alongside Scout/Maverick, still in training at launch. Tops benchmarks in STEM reasoning.
Capabilities and innovations
2T total / 288B active MoE16 experts, 4 activeSTEM reasoningNatively multimodalIn training at announcementLargest Llama Model EverTeacher Model Architecture288B Active Parameters
Benchmarks
No benchmark values are recorded for this model.
- GPQA Diamond: not reported by the vendor
- Humanity’s Last Exam (no tools): not reported by the vendor
- SWE-bench Verified: not reported by the vendor
- SWE-bench Pro: benchmark did not exist at release
- MMLU: not reported by the vendor
- MMLU-Pro: not reported by the vendor
- Terminal-Bench 2.x: benchmark did not exist at release
- MMMU-Pro: benchmark did not exist at release
Architecture and hardware
- Parameters
- 2000B total, 288B active per token (MoE)
- Estimated VRAM at Q4
- ~1150 GB, Frontier class
Links
More from Meta
Data curated by AI Model Timeline. See the methodology for admission criteria, benchmark eras and source priority.