Models / Clef-Omni

Clef-Omni

MoE enthusiast
HuggingFace → Cloudflare → Find for my hardware

The Clef family grows an omni head. Cloudflare’s Clef-Omni (October 9, 2026) is a 35.3B omni model on the Qwen3-OMNI MoE architecture: audio in (and out via the talker), image in, text out, Apache 2.0. Same decision-model contract as its 27B and 9B siblings: map an input to action probabilities, never write an essay.

Fit math. The thinker runs 48 layers at 4 KV heads x 128 head-dim to a 65,536 context: about 96MB of cache per 1,000 tokens and about 20.2GB of q4 weights for the full omni stack. A 24GB to 32GB card hosts the whole thing; the flash-tier split (text decision vs omni IO) is what keeps the small builds cheap.

Why Cloudflare keeps shipping these. Three Clef-class models in nine days (27B, 9B flash, omni) is an edge-infrastructure strategy, not a model line: a CDN whose Workers run customers’ routing decisions wants a model family sized for edge GPUs, and open weights are how it recruits the ecosystem to build on that family.

decision-model omni audio vision routing
Parameters
35.3B
Context
65k
License
apache 2.0
Developer
Cloudflare
Origin
🇺🇸 USA
Released
Oct 2026

Related models

Clef
27.4B enthusiast 16GB min RAM Oct 1, 2026
Clef-flash
9.4B consumer 8GB min RAM Oct 1, 2026

Save your hardware and every model page answers the real question: will it run on your machine, and how fast?

Join free - save your rig →

Run it locally

Per-quant memory needs and a static "can you run it?" reference - no rig entry required

Q4_K_M
20.2GB 24.0GB min 32.0GB rec
Balanced - the usual local sweet spot

The reference hardware

Schematic of the NVIDIA Jetson Orin NX 16GB reference rig - 16GB unified memory, 102 GB/s aggregate bandwidth
Schematic of the Single GTX 1080 Ti (11GB) reference rig - 11GB VRAM, 484 GB/s aggregate bandwidth
Schematic of the Single RTX 4090 (24GB) reference rig - 24GB VRAM, 1008 GB/s aggregate bandwidth
Schematic of the 4x H100 80GB (320GB) reference rig - 320GB VRAM, 13400 GB/s aggregate bandwidth
Schematic of the NVIDIA DGX Station 748GB reference rig - 748GB unified memory, 8000 GB/s aggregate bandwidth
Schematic of the 8x RTX 3090 rack (192GB) reference rig - 192GB VRAM, 7489 GB/s aggregate bandwidth
Schematic of the 4x RTX 5090 (128GB) reference rig - 128GB VRAM, 7168 GB/s aggregate bandwidth
Schematic of the AMD Instinct MI300X (192GB) reference rig - 192GB VRAM, 5324 GB/s aggregate bandwidth
Schematic of the 4x RTX 4090 (96GB) reference rig - 96GB VRAM, 4032 GB/s aggregate bandwidth
Schematic of the 2x RTX 5090 (64GB) reference rig - 64GB VRAM, 3584 GB/s aggregate bandwidth
Schematic of the 2x RTX 3090 (48GB) reference rig - 48GB VRAM, 1872 GB/s aggregate bandwidth
Schematic of the Single RTX 5090 (32GB) reference rig - 32GB VRAM, 1792 GB/s aggregate bandwidth
Schematic of the RTX PRO 6000 Blackwell (96GB) reference rig - 96GB VRAM, 1792 GB/s aggregate bandwidth
Schematic of the Mac Studio M4 Ultra 192GB reference rig - 192GB unified memory, 1092 GB/s aggregate bandwidth
Schematic of the Mac Studio M4 Ultra 512GB reference rig - 512GB unified memory, 1092 GB/s aggregate bandwidth
Schematic of the MacBook Pro M5 Max 128GB reference rig - 128GB unified memory, 614 GB/s aggregate bandwidth
Schematic of the Dual EPYC 9004 + 768GB DDR5-4800 reference rig - 768GB unified memory, 460 GB/s aggregate bandwidth
Schematic of the DGX Spark 128GB unified reference rig - 128GB unified memory, 273 GB/s aggregate bandwidth
Schematic of the Ryzen AI Max+ 395 128GB reference rig - 128GB unified memory, 256 GB/s aggregate bandwidth
Schematic of the Jetson AGX Orin 64GB reference rig - 64GB unified memory, 204 GB/s aggregate bandwidth
Schematic of the Epyc + 512GB DDR4-3200 + 2x RTX 3090 reference rig - 560GB unified memory, 204 GB/s aggregate bandwidth
Schematic of the Epyc + 512GB DDR4-2400 + 2x RTX 3090 reference rig - 560GB unified memory, 153 GB/s aggregate bandwidth

22 reference configs, drawn in-house. Scroll for more.

Can you run it? - reference rigs

Rig Q4_K_M
NVIDIA Jetson Orin NX 16GB no -> cloud
Single GTX 1080 Ti (11GB) no -> cloud
Single RTX 4090 (24GB) tight
4x H100 80GB (320GB) fast 3505.1t/s
NVIDIA DGX Station 748GB fast 2092.6t/s
8x RTX 3090 rack (192GB) fast 1959.1t/s
4x RTX 5090 (128GB) fast 1875.0t/s
AMD Instinct MI300X (192GB) fast 1392.8t/s
4x RTX 4090 (96GB) fast 1054.7t/s
2x RTX 5090 (64GB) fast 937.5t/s
2x RTX 3090 (48GB) fast 489.8t/s
Single RTX 5090 (32GB) fast 468.7t/s
RTX PRO 6000 Blackwell (96GB) fast 468.7t/s
Mac Studio M4 Ultra 192GB fast 311.6t/s
Mac Studio M4 Ultra 512GB fast 311.6t/s
MacBook Pro M5 Max 128GB fast 175.2t/s
Dual EPYC 9004 + 768GB DDR5-4800 fast 120.5t/s
DGX Spark 128GB unified fast 71.4t/s
Ryzen AI Max+ 395 128GB fast 67.0t/s
Jetson AGX Orin 64GB fast 53.6t/s
Epyc + 512GB DDR4-3200 + 2x RTX 3090 fast 53.6t/s
Epyc + 512GB DDR4-2400 + 2x RTX 3090 fast 40.2t/s

Fit tiers use the same will-it-run logic as the rig finder. For comfortable fits, the badge reflects decode speed: fast >=20 t/s, ok 8-20 t/s, slow <8 t/s. t/s is a bandwidth estimate, not a measured benchmark.

Download options

Q4_K_M community
20.2GB dl 24.0GB min 32.0GB rec
REC RAM vs largest quant
20.2GB q4 full omni stack + KV 96MB/1k (thinker 48L x 4 kv x 128); 65k ctx at 6GB cache
GGUF on HF →

Or run it in the cloud

No per-token API provider pricing tracked for Clef-Omni yet. For flagship list prices, see the calculator.

PRICE HISTORY

Inference cost over time

Data accumulates from the first daily sync - longer ranges populate over time. Prices come from OpenRouter snapshots, not a historical API.

Loading price history...