// Performance

Benchmarks

Uneven AI is engineered with a high-performance native core. Below are the actual verified benchmarks obtained from running Uneven in our testing environment. By showing complete execution environments, versions, and system constraints, we aim to be fully transparent about our performance claims.

Hardware & Testing Environment

All benchmarks were executed on a low-spec host machine to verify resource constraints, peak load handling, and execution efficiency under strict resource limitations.

Hardware
Positivo Informatica SA S14BW01
Processor
Intel® Pentium® N3710 (4 Cores, 1.60GHz)
RAM
8.0 GiB (7.1 GiB usable)
OS / Kernel
Ubuntu 26.04 LTS | Linux 7.0.0-22-generic
Disk
240.1 GB SATA SSD

Verified Benchmark Runs

Each card below represents a single benchmark report run, detailing the version used, operation type, resources consumed, and links to raw monitoring outputs.

v1.12.2 (release) Success

RAG Context Retrieval & Q&A

Queries the model with the prompt: "How does the scheduler balance tasks across CPU cores?", dynamically loading 8 relevant kernel scheduler files. Optimized in Release mode.

Scope:46 files (120,685 tokens)
Engine/Model:gemini-2.5-flash (API)
Processor:Intel® Pentium® N3710 (4 Cores, 1.60GHz)
OS:Ubuntu 26.04 LTS | Linux 7.0.0-22-generic
29s
Duration
120 MB
Peak Memory (RSS)
600%
Peak CPU
v1.12.1 (debug) Success

RAG Context Retrieval & Q&A

Queries the model with the prompt: "How does the scheduler balance tasks across CPU cores?", dynamically loading 8 relevant kernel scheduler files. Run in Debug mode.

Scope:46 files (121,330 tokens)
Engine/Model:gemini-2.5-flash (API)
Processor:Intel® Pentium® N3710 (4 Cores, 1.60GHz)
OS:Ubuntu 26.04 LTS | Linux 7.0.0-22-generic
34s
Duration
149 MB
Peak Memory (RSS)
250%
Peak CPU
v1.12.2 (release) Success

Subsystem Workspace Indexing

Performs a clean, full index build of the scheduler subsystem source files (hashing and mapping content). Optimized in Release mode.

Scope:46 files (13.3 KB total)
Engine/Model:Offline (Native Core)
Processor:Intel® Pentium® N3710 (4 Cores, 1.60GHz)
OS:Ubuntu 26.04 LTS | Linux 7.0.0-22-generic
8s (7.31s native)
Duration
24 MB
Peak Memory (RSS)
500%
Peak CPU
v1.12.1 (debug) Success

Large-Scale Workspace Indexing

Builds incremental index metadata for the full Linux Kernel Stable repository. Run in Debug mode.

Scope:93,441 files (2.2 GB total)
Engine/Model:Offline (Native Core)
Processor:Intel® Pentium® N3710 (4 Cores, 1.60GHz)
OS:Ubuntu 26.04 LTS | Linux 7.0.0-22-generic
6m 20s (378.68s)
Duration
155 MB
Peak Memory (RSS)
500%
Peak CPU

Comparative Visualizations

Below are comparative visualizations based on our testing data, demonstrating memory usage and token economy.

Memory Allocation (Peak RSS) by Workspace Size
Scheduler Subsystem Indexing (46 files - v1.12.2 Release)
24 MB
Full Linux Kernel Indexing (93,441 files - v1.12.1 Debug)
155 MB
Token Billing Costs on Massive Q&A (Cold vs Warm Cache)
Standard/Cold Cache Query Cost
$1.88 / Query
Uneven AI (Gemini 2.5 Flash Warm Cache)
$0.037 / Query