Are you absolutely sure?

ALL-TIME TOKENS
10.7M

SINCE NETWORK GENESIS

TRAILING 24 HOURS
3.5M

WINDOW ENDING AT CAPTURE

PUBLIC NETWORK TELEMETRY

NETWORK
PULSE

All-time, network-wide analytics of the Cestus inference network. Aggregated daily, free of user and node identifiers.

01 TOKEN TREND

Daily network tokens over the trailing 90 days, UTC. Top models drawn against the network total; the strip above carries the all-time counters.

TOTAL_REQUESTS
1,851

SINCE GENESIS

REQUESTS_24H
502

TRAILING WINDOW

IMAGES
148

IMAGES GENERATED

VIDEOS
47

3m 29s

ENERGY
6.9 kWh

ESTIMATED DRAW

[ DAILY_TOKENS ] TRAILING 90 DAYS · UTC
NETWORK_TOTAL qwen3.6-35b glm-4.7-flash gemma-4-26B qwen3.6-27b-uncensored
4.8M 31 Jul
INPUT
3.3M
OUTPUT
7.4M
CACHED
1.6M
CACHE_HIT_RATE
50.1%

02 MODEL DEMAND

Text models ranked over successful requests in the published snapshot, with the network total closing the table.

[ TEXT_MODELS ] ALL-TIME · RANKED BY TOTAL TOKENS
MODEL INOUTTOTALCACHEDRATIOREQ
01 qwen3.6-35b 2.6M 7.0M 9.6M 1.3M 48.6% 1,367
02 glm-4.7-flash 600.9K 320.4K 921.2K 362.7K 60.4% 181
03 gemma-4-26B 31.8K 65.6K 97.5K 528 1.7% 115
04 qwen3.6-27b-uncensored 9.0K 38.1K 47.1K 3.8K 42.0% 39
05 gemma-4-12b 17.0K 15.4K 32.4K 1.3K 7.6% 15
06 muse-glimmer-30b 197 94 291 0 0.0% 2
NETWORK TOTAL 3.3M 7.4M 10.7M 1.6M 50.1% 1,719

[ IMAGE_MODELS ] ALL-TIME

  1. 01 sd-1.5

    109 IMAGES

    46 REQ 73.6%

  2. 02 flux2-klein-9b

    39 IMAGES

    39 REQ 26.4%

  3. ΣALL MODELS

    148 IMAGES

    85 REQ100%

03 · NON-TOKEN MODALITIES

SYNTHESIS

SYS_PHASE_03 ACTIVE

NATIVE OUTPUT UNITS

Image and video work carries no tokens. Both are counted, ranked and shared in the unit they actually produce.

IMAGES
148
CLIPS
47
RUNTIME
3m 29s

[ VIDEO_MODELS ] ALL-TIME

  1. 01 minimax-h3

    38 CLIPS

    3m 11s 80.9%

  2. 02 wan2.1-t2v-1.3b

    5 CLIPS

    10.3s 10.6%

  3. 03 wan2.2-ti2v-5b

    4 CLIPS

    7.6s 8.5%

  4. ΣALL MODELS

    47 CLIPS

    3m 29s100%

04 GPU CAPACITY

10.6M ATTRIBUTED TOKENS

Online units come from the live node inventory; tokens are attributed execution across the snapshot. Multi-GPU nodes split their tokens evenly. 0 sit outside the ranking: requests without a node or with unrecognised hardware still count toward every total above.

[ TOKENS_BY_GPU ] RANKED BY ATTRIBUTED TOKENS
  1. 01 RTX 4090 2 81.5% 8.7M
  2. 02 RTX 3090 - 18.5% 2.0M