Skip to content

RunInfraby RightNow

Model APIsNew
Pricing
Research
Contact

Dashboard Sign in Get started

Measured optimization techniques for open model inference

Loading optimization

Optimization

Measured optimization techniques for open model inference

The published catalog currently carries 2 measured techniques across 3 of 4 published packages.

Channelwise FP8 weights, dynamic per-token activations
2 published packages
Measured models
AREX-Turbo and Qwythos-9B-Claude-Mythos-5-1M
Accuracy recovery
Lowest measured recovery 99.35%. Significance is not established; within-noise changes are not accuracy gains.
Raw score ratios
99.35% to 107.14%. These ratios alone do not establish accuracy gains.
Channelwise FP8, measured selective-layer recipe
1 published package
Measured models
Qwen3.6 27B
Accuracy recovery
Lowest measured recovery 99.87%. Significance is not established; within-noise changes are not accuracy gains.
Raw score ratios
99.87% to 99.87%. These ratios alone do not establish accuracy gains.

Call Nemotron 3.5 Lightning 30B with Model APIs

Use a workspace API key and pay for input, cached input, and output tokens.

View Model APIs

TLS in transit, AES-256 at rest

Workspace isolation

No training on your data

SOC 2 Type II

RunInfraby RightNow

© 2026 RunInfra. All rights reserved.

Join the community System status

Coding plans Model APIs Pricing Startups Benchmarks Docs Research News Brand assets Contact

Backed by

Y Combinator

Ask AI about RunInfra

Part of RightNow

Security DPA AUP Cookies Terms Privacy