Home / Models / Agnes 3.0 Flash

How to Run Agnes 3.0 Flash Locally — Setup Guide (September 2026)

Agnes 3.0 Flash Preview is a 33B dense multimodal model from Agnes AI, released under Apache-2.0. Three of every four layers are recurrent gated-delta layers that carry a fixed-size state, so only 18 of its 72 layers keep a KV cache that grows with context. The GGUFs are a conversion onto llama.cpp's existing Qwen3.5 graph, which means it runs on a recent stock llama.cpp build.

Numbers you can trust

We don't print guessed speeds. Every measured number on this page came from TurboLLM's own auto-tuner running Agnes 3.0 Flash on real hardware — labeled as such. When you run it yourself, TurboLLM benchmarks it on your exact GPU and shows a VRAM-fit verdict before you load, and the real measured tokens/sec once it's running.

Does it fit your GPU?

33.1B dense parameters, Apache-2.0, 72 layers: 54 gated-delta (recurrent) layers plus 18 global-attention layers (4 KV heads, 256 head dimension), with a 262K native context and a vision tower. This is the earlier open-weight Preview checkpoint — Agnes's hosted API model is a different checkpoint. The GGUFs are community conversions, and the multi-token-prediction weights are left out.

Agnes 3.0 Flash Preview
Agnes AI · GGUF
Dense · hybrid gated-delta attention · vision
visionthinking
VRAMFit
24 GBQ4_K_M (19.8 GB) fully in VRAM with room for a long context, or Q5_K_M (23.0 GB) at a shorter one · 262K native context
32 GBQ6_K (26.4 GB) fully in VRAM · 262K native context
48 GB+Q8_0 (34.2 GB) fully in VRAM · 262K native context
0xKitkat/Agnes-3.0-Flash-GGUF
View on Hugging Face →

How to run it

  1. Install TurboLLM

    npx turbollm — no install step, works on Windows, macOS, and Linux. It detects your GPU and auto-provisions a matching llama-server build (no CUDA toolkit, no Python, no compiler).

  2. Get the model

    Paste 0xKitkat/Agnes-3.0-Flash-GGUF into TurboLLM's in-app Hugging Face search — it lists every quant with a VRAM-fit verdict against your real free VRAM, and downloads are resumable and SHA-256 verified.

  3. Let auto-fit pick the setup

    Flip the auto-fit toggle and TurboLLM chooses the GPU/CPU layer split (and, for a Mixture-of-Experts model, the expert-offload split) for your card — the same manual flag-hunting every other guide walks you through by hand.

  4. Load it

    TurboLLM benchmarks the model on your exact GPU as it loads and shows the real measured tokens/sec — not a number copied from someone else's card.

  5. Use it

    Chat in the built-in UI, or point any OpenAI- or Anthropic-compatible tool (including Claude Code) at TurboLLM's local API — see the API docs.

One honest caveat

The GGUF author validated this at a 4K context with thinking off. Long context, reasoning-on quality, tool calling and video were not tested. Vision needs the separate mmproj-Agnes-3.0-Flash-F16 file (0.9 GB) from the same repo. Stock llama.cpp does not apply the upstream tokenizer's NFC Unicode normalisation, so text with decomposed accents can tokenise differently from the reference.

FAQ

Will Agnes 3.0 Flash run on my GPU?

Check the VRAM table above for the tier closest to your card. If you're not sure which tier you're in, or want picks tailored to your exact hardware, use What can I run? — enter your VRAM (or Apple unified memory) and it suggests the model, quant, and offload split that fits, the same logic TurboLLM's auto-fit runs when you load a model.

Which quant should I use for Agnes 3.0 Flash?

Pick the highest quant that fits your VRAM with room left over for the KV cache and compute buffer — not just the model weights. As a starting point, Q4_K_M is usually the sweet spot (small quality loss for a large size drop); step up to Q5_K_M/Q6_K/Q8_0 if you have headroom, or down to Q3_K_M/an IQ quant if you don't. TurboLLM shows a VRAM-fit verdict for every quant against your card's real free VRAM before you download — see Quantization explained for what the letters and numbers actually mean, or llama.cpp flags that actually matter for the other settings worth knowing.

Why does Agnes 3.0 Flash suddenly get slow or stutter?

The usual cause is silent VRAM spill: once a model's weights plus its KV cache stop fitting in VRAM, the driver quietly offloads the overflow to system RAM over PCIe, and that path runs 5–10× slower with no error message — just a sudden slowdown. TurboLLM's auto-fit sizes the quant and GPU/CPU split against your card's real free VRAM (including KV-cache growth at your chosen context) specifically to avoid this, and shows a VRAM-fit verdict before you load rather than after it's already crawling. Full mechanics in VRAM spill explained.

Run it now

$ npx turbollm

One command detects your GPU, provisions the right engine, and opens the UI. New here? Start with Install & first run and Quantization explained. On different hardware, see RTX 3090, RTX 4090 or RTX 5080. Also see how to run it or how to run it.