
AgentX is InferenceX's new benchmark for agentic inference performance from SemiAnalysis.
The Hook
“(No spoken dialogue — ambient electronic synth music)”
Problem Agitation
“We now have machines that can think, but how do we measure their performance?”
Product Reveal
“InferenceX is the definitive benchmark for AI inference performance. The new agentic inference benchmark showcases real-world chip performance across many of the most popular frontier open source models.”
Feature Teaser
“Select from a wide range of hardware and inference engine configurations to view and compare their performance frontiers. Choose between common latency metrics such as interactivity, time to first token, and end-to-end latency.”
Feature Teaser
“Select a y-axis metric to compare what matters most to you. Track a chip configuration's performance over time to see how inference optimizations drive measurable performance gains.”
Feature Teaser
“Each point on the Pareto frontier represents a single benchmark run. Hover over any point to explore detailed performance data. Click View Charts to view server metrics: throughput, KV cache metrics such as token source, KV working set size, prefix cache hit rate, and utilization over time.”
Feature Teaser
“View the actual timeline of requests received by the server, including realistic sub-agent rollouts. Click any request to inspect its source and token breakdown from the underlying dataset. Inspect logs from every part of the benchmark, including servers, routers, KV transfer engines, and more.”
Feature Teaser
“Each benchmark runs on real hardware, managed by SemiAnalysis through self-hosted GitHub Actions, with the full run linked for every data point.”
Call to Action
“InferenceX is 100% open source. Every result is reproducible, no private hyper-optimized containers. InferenceX is the authoritative index for real-world inference performance. Check it out today for yourself at inferencex.com.”
Synthesizable in Remotion · key components of the developer-benchmark-showcase template.
Draws SVG scatter dots and smooth connecting trade-off lines dynamically across X/Y axis transitions.
Use interpolate() along strokeDashoffset for path animations and spring() on circle scales based on frame.
Renders horizontal execution bars sequentially with hover inspection details and latency breakdown badges.
Map over request timestamps, interpolate width and translateX per turn using Remotion spring physics.
Types white monospaced characters on center screen with a blinking block cursor.
Calculate slice of text string based on frame count with an opacity toggling span at 30fps.
Purpose. Ground modern LLM workloads in hardware history.
Execution. Rapid die-shot montage terminating on AI server dies.
Purpose. Establish lack of realistic AI inference benchmarking.
Execution. Terminal logs and large typographic challenge question.
Purpose. Showcase multidimensional Pareto evaluation platform.
Execution. Live app screencasts switching architectures, models, and cost axes.
Purpose. Validate technical depth with low-level server metrics.
Execution. Full-bleed dashboard graphs, timeline waterfall, and raw logs.
Purpose. Confirm open-source validity and direct viewer to website.
Execution. GitHub Actions execution view followed by homepage quote cards and URL.
Objective
Launch video for an open-source database latency benchmarkHow the recipe adapts
Replace GPU dies with storage drive PCBs, substitute model configs with SQL storage engines, and render read/write throughput versus P99 latency frontiers.Scene durations (7 scenes, 139s)
Key Remotion components
To ensure that artificial general intelligence benefits all of humanity

Gopuff introduces Go, an AI shopping assistant built with SpaceXAI: say what you need and the order is placed.

Iliad (Troy) trailer made by Grok Imagine 1.5, which was just released