Best local LLM for roleplay & writing on a Apple M3 Ultra (96GB) (2026)
All figures assume an f16 KV cache, a 0.6 GB display reserve on the GPU, and 64 GB of DDR5 system RAM for the offload tiers. Tune these in the calculator.
The verdict
Rocinante 12B Q8_0
Fits on GPU at 16K · Roleplay & writing score 9/10 · ≈ 40 tok/s
The budget roleplay king. Lowest slop-per-token of anything under 24 GB; the community keeps it alive for a reason.
llama-server -m Rocinante-12B-v1.1-Q8_0.gguf -c 16384 --flash-attn -ngl 99
Worthy alternates
Cydonia 24B Q8_0
Fits on GPU · ≈ 21 tok/s · Roleplay & writing 9/10
Rocinante's bigger sibling on a Mistral Small base. The default serious-RP pick for 24 GB cards.
Anubis 70B Q6_K
Fits on GPU · ≈ 9 tok/s · Roleplay & writing 9/10
The 70B-class roleplay tune with current GGUFs. Llama 3.3 prose depth, none of the corporate manners.
Tune this for your exact RAM and settings in the calculator → · All models on the Apple M3 Ultra (96GB)
Frequently asked questions
What is the best local LLM for roleplay & writing on a Apple M3 Ultra (96GB)?
Rocinante 12B at Q8_0 — it scores 9/10 for roleplay & writing and runs as "Fits on GPU" at 16K context on the Apple M3 Ultra (96GB).
How much context do I need for roleplay & writing?
We recommend 16K tokens for roleplay & writing (minimum 8K). These picks are computed at 16K.
How fast will it run on a Apple M3 Ultra (96GB)?
Roughly 40 tokens/sec for Rocinante 12B — comfortable for interactive use.
Do I need more than 72 GB of VRAM for roleplay & writing?
No — the pick above needs 16.5 GB of VRAM at 16K.
What settings should I use?
Start with our command: llama-server -m Rocinante-12B-v1.1-Q8_0.gguf -c 16384 --flash-attn -ngl 99 — then tune context and KV quant in the fit calculator.