Performance lab


A token-per-second number is not a system benchmark.

Users feel time to first token, reading cadence and queueing. Operators feel aggregate throughput, tail latency, energy and failures. Compliance teams need quality and refusal evidence. We measure all three views.

Measured results onlyMethod aligned to public standards

Build the benchmark contract

Generate the scenario that belongs in a proposal and acceptance test. This creates a protocol, not a performance estimate.

Local protocol tool

No prompt, document or selection leaves this page.

Protocol output

Choose a system and workload.

Public result status

We would rather publish an empty measured table than fill it with vendor numbers labelled as ours.

ANULUM measured
0production configurations published
7standard acceptance scenarios defined
p50 + p95latency distribution required
raw JSONrequired with every result

Current boundary: no absolute ANULUM throughput result is published yet. Vendor claims may inform planning, but they never enter the green measured class.

Seven acceptance scenarios

Interactive C12K in · 512 out · one user
Interactive C42K in · 512 out · four users
Interactive C82K in · 512 out · eight users
Document 32K32K in · 512 out · cited answer
Long context 128K128K in · 1K out · one user
RAG-2020 chunks · answer + citations
Agent-1010 tool steps · deterministic task

What every result must disclose

Exact artefact

Model revision and hash, licence, quantisation, serving engine and version.

Exact machine

Accelerators, memory, CPU, RAM, fabric, power limits, software and driver versions.

Exact workload

Input and output length, concurrency, cold/warm state, sampling and repetition count.

User latency

TTFT p50/p95 and inter-token latency p50/p95 — not averages alone.

Operator capacity

Per-request and aggregate output tokens/s, queueing, failures and energy per 1,000 output tokens.

Answer quality

Groundedness, citation correctness, unsupported-claim refusal and access-control tests.

Evidence legend

ClassMeaningMay support a client promise?
OfficialA source-controlled vendor or model-publisher specification.Only for the specification it states.
MeasuredANULUM result with reproducible configuration and raw evidence.Yes, within the tested envelope.
CalculatedCapacity or cost derived from explicit assumptions.Planning only.
ProvisionalCommercial or technical value awaiting quotation or validation.No.

MLPerf Inference documentation ↗

Put acceptance criteria in the contract.

The architecture review defines which scenarios, latency limits, quality gates and evidence artefacts govern procurement.

Define acceptance criteria See engagement bands →