AI Workload Estimator · University of Idaho

Plan an LLM-assisted research project before you start

Have an AI read 40,000 survey comments or 2 million abstracts? Estimate the tokens, wall-clock time and quota you will need on MindRouter — and pick the right model — before you commit a single request.

Corpus

Number of items

Survey comments, abstracts, transcripts, code files…

Average length per item

1 page ≈ 300 words ≈ 390 tokens

Paste a real sample (optional, more accurate)

No sample yet — using the length above.

Task & model

Task template

Model

Throughput assumptions

estimates

MindRouter's model catalog does not include throughput, pricing or quota figures. These are editable estimates — set them to match what you observe on MindRouter and your concurrency.

Tokens per second, by model class

Class	Prompt tok/s	Generation tok/s
Small ≤ 14B
Medium 15–40B
Large > 40B
Embedding

Prompt (prefill) tokens are processed far faster than generated tokens. Each model's class is derived from its id.

Parallel requests concurrency

Assumes each request keeps its speed as concurrency rises — real servers slow down under heavy load.

Per-request overhead seconds

Prompt / instruction overhead tokens per call

Safety margin tokens per call

Results

estimates

Describe your corpus to see an estimate

Enter a number of items and an average length (or paste a sample) in step 1, then pick a task and a model. Everything below updates live.

Quota

your own budget — quotas vary by group, so enter yours

Token budget per period (tokens)

= 100M tokens (placeholder — enter your own)

Requests per minute limit

Budget period

Timeline

Sensitivity

drag to see how time & tokens move

Parallel requests4

Generation speed1.0×

Items40,000

Output length1.0×

Compare 3 models

pick any three — click a row to select it

Model	Class	Window	Fits?	Tokens	Time

Chunking visualizer

Chunk overlap tokens

Python batching snippet

OpenAI-compatible MindRouter endpoint, your settings

Data & sources

MindRouter model catalog — /v1/models, University of Idaho Research Computing and Data Services (RCDS) / Institute for Interdisciplinary Data Sciences (IIDS). Snapshot retrieved 2026-09-13: 73 entries (63 models + 10 aliases). mindrouter.uidaho.edu The raw endpoint requires a MindRouter API key, so we link the public home page.
Throughput, time & quota figures — not from MindRouter. The catalog contains no tokens/second, latency, pricing or quota data. Every time/throughput number in this app is an editable estimate you set in step 3; nothing here is a real University of Idaho figure.
Token counts — computed with an approximate JavaScript tokenizer (word / sub-word / non-Latin heuristics). These are estimates of fit, not exact counts from each model's own tokenizer.