Free, open-source alpha. Benchmarks are real, measured in Claude Opus 5.5 test runs. Less data, less energy: nocap →

khėnsu.ñü

nocap · capitalism to postcap · free, open-source alpha

less data. less energy.
more left over for everyone.

  • 41–66 %: Fewer bits than H.264 for the same picture, measured on an iPhone Air
  • 1.7–2.9 ×: Viewers served on the same delivery bill
  • 0: Server-side AI calls: assistant and memory run on the device
  • $0: Licence fee. Free and open source, alpha

Benchmarks are real: measured in Claude Opus 5.5 test runs on an iPhone Air (iPhone18,4) and a Mac with M1 Pro, scored with Netflix’s open VMAF. Everything else on this page is a projection from those benchmarks, with the inputs shown so you can change them.

AI datacenter emissionsWhat it means for a businessReview the code and docs

greenhouse gas

cutting AI datacenter emissions at the source.

The cheapest kilowatt-hour is the one a datacenter never draws. khėnsu.ñü moves work off the datacenter in three ways, and every one of them is in the alpha.

Usual pathPhoneDatacenterAI + full videoViewerkhėnsu.ñüPhoneAI on deviceMixnet relayforwards bitsViewerrebuilds the picture
Thick lines: full-size media and AI requests through a datacenter. Thin line: the base layer only; the picture and the AI run on the devices at each end.
  1. Inference on the device.Ask khėnsu, nu.ral memory and the summaries run on the Neural Engine of the phone or Mac they serve. No Siri, no Private Cloud Compute, no server model, so a question costs a datacenter nothing.
  2. Send less, rebuild at the edge.ñü.media sends a small hardware-encoded base layer and the receiving device rebuilds the full picture. Datacenters, CDNs and networks carry 41–66% fewer bits than H.264 for the same VMAF.
  3. Hardware media engines, not GPU software.The base layer is the phone’s own HEVC encoder, in real time. Measured on the iPhone Air, it matches real-time AV1 software at 720p and beats it at 4K, without a GPU in a rack doing the encode.

—MWh a year not drawn

—tonnes CO₂e a year avoided

Projection. Energy per GB for networks and datacenters varies tenfold between published studies, so set your own figure; the default is an example, not a measurement. Bits saved is the measured range against H.264.

thank you

built with claude opus 5.5 and sol astra.

anthropic · claude opus 5.5

Opus 5.5 wrote most of this code and ran every benchmark on this page, on real devices, and its numbers held up. Thank you to Anthropic and its founders, Dario Amodei, Daniela Amodei, Jared Kaplan, Sam McCandlish, Tom Brown, Chris Olah, Jack Clark and Ben Mann, for building a model careful enough to trust with the measurements.

sol astra

Sol Astra has been invaluable to this work alongside Opus. Thank you to the company and the founders who built it. Two very different minds made the work better than either alone.

capitalism to postcap

same picture. 41–66% fewer bits.

Bits needed for the same VMAF score, measured on an iPhone Air with whole clips encoded over USB. Against HEVC the gain is 7–16%; most services still ship H.264.

720p reelInstagram reel, 720 × 1280, 30 fps · same picture: VMAF 75
  • H.2641.19 Mbit/s
  • HEVC755 kbit/s
  • ñü.media702 kbit/s

41% fewer bits than H.264 · 7% fewer than HEVC

benchmark
4K auroraNASA aurora from the ISS, 3840 × 2160, 30 fps · same picture: VMAF 57
  • H.2642.76 Mbit/s
  • HEVC1.13 Mbit/s
  • ñü.media949 kbit/s

66% fewer bits than H.264 · 16% fewer than HEVC

benchmark
Quality against bitrate, 720p reelVMAF, 0 to 100, against the original. Higher and further left is better.
020406080100125k250k500k1M2M
  • ñü.media
  • HEVC, iPhone hardware
  • H.264, iPhone hardware
benchmark

lower bills

CDN egress, transit and storage are priced by the byte. 41–66% fewer bytes is 41–66% off those lines, or 1.7–2.9× the viewers on the same bill.

faster services

Fewer bits start sooner and stall less on the same connection, and the base layer encodes on the phone’s hardware in real time, so live video goes out without a transcoding farm.

lower emissions

Every byte not sent is energy not drawn in datacenters, networks and radios. Pair it with on-device AI and the whole session leans on the devices people already own.

how much more could you keep?

—a month off delivery

—more profit

Projection. With the example inputs, a delivery-heavy service keeps 30–40% more profit at the low end of the measured savings. Put in your own numbers; the only measured input is the bits saved.

moqom.cloud

free MoQ, on cloudflare’s MoQ relay beta.

moqom.cloud gives you Media over QUIC for free, with open SDKs for Swift, Go and Rust, running on the MoQ relays Cloudflare made available in beta on 11 October 2026 and free during the beta. Ship ñü.media over MoQ and the savings above reach your viewers with sub-second latency.

Get free MoQ on moqom.cloudCloudflare MoQ docs

moqom.cloud is an independent project, not affiliated with or endorsed by Cloudflare.

review and assess

check every number yourself.

This is an alpha. Read the design, rerun the benchmarks with public tools, and tell us what breaks.