Onboarding

Welcome. Let's get you started.

LLM evaluation and load testing, run for you. Pick what you want to test, choose a size and an area, and we do the rest. One price, with token and GPU spend included.

Frontier models

Which model to use for which task and sub-task.

See how it works →

Frontier models, enhanced

RV_Enhanced: a team of frontier models working together, for hard problems and when accuracy is important.

See RV_Enhanced →

Open models, your endpoint

Which open model to use, and how much load your endpoint can take.

Start with your endpoint →

Open models, your server

SGLang, vLLM or TensorRT-LLM, and the settings to use on your GPUs.

Start with your server →

Ready to onboard? Tell us what you are interested in.

Frontier models

You choose a size and an area. We do the rest.

STEP 1

Place an order.

Pick a size. It sets the number of questions and how hard they are. Every size has at least 200 questions, so that the tests can separate the models.

Core200 questions, standard complexity.
Advanced400 questions, harder.
Expert1,000 questions, hardest.

These counts are representative. We set the exact number for your tasks and sub-tasks. One price, with token spend included.

STEP 2

Build the questions.

There are four ways to create the question set. Choosing your areas is all you have to do. The other three take optional input from you, for when you want a say. You do not have to provide the questions: if the price works for you, we write them all.

Start here

Choose your areas

Pick as many as you like. Each tag is an area or a sub-area, and two related ones work well together.

Filled by RadianVector: The questions, the answers, and how to check them.
Optionally checked by you: All of it, at sign-off.

Optional input from you

Let us look at your product

Filled by RadianVector: The questions, from your site and docs. The answers, and how to check them.
Optionally checked by you: All of it, at sign-off.

Optional input from you

Describe your tasks

Filled by RadianVector: The questions for those tasks, and their answers. We decide how to check each answer.
Optionally checked by you: You can tell us how to check, if you prefer.

Optional input from you

Send your questions or traces

Filled by RadianVector: Any answers you leave empty, and how to check them.
Optionally checked by you: You can flag the answers that matter, or tell us how to check them.

The format: a CSV file

One row per question, for anything you send us. Send as many as you have, and we write the rest. Only question is required. If you leave the answer and how_to_check fields empty, we will fill them in on our end. We send the question set back in the same format.

Download the template

questions.csv
question,answer,how_to_check,area,sub_area
What is the total federal tax?,122175.83,"A number, within $2",Taxes,Investor return
Which clause caps liability here?,,Must name the clause,Law,Contracts
What dose for a 72 kg patient at 5 mg/kg?,360 mg,,Healthcare,Dosing

STEP 3 · OPTIONAL INPUT FROM YOU

Sign off, or let our systems decide.

You can read every question with its answer, change a group, and sign off. Or let our systems decide: the run starts once the set has at least 200 questions. If you sign, we record exactly which version you approved.

STEP 4 · NO INPUT FROM YOU

We run them.

We run every question on every viable combination of model, effort, system prompt and output format. If we know a combination will do badly, we leave it out. The rest include system prompts tuned with DSPy, GEPA and MIPROv2, and replies as JSON, as plain text and with the reasoning first. What the research shows.

Why we vary these: what the research shows.

STEP 5 · NO INPUT FROM YOU

Get the report.

We grade every answer, then run the answers through our repos, shown here as rv-1 to rv-n, and our statistical tests, including McNemar, Holm and repetition tests. You get correctness, speed and token spend, and a report of which model to use for which task and sub-task.

Which model for which sub-task

Sub-task	Claude Sonnet	GPT-6 Astra	Grok 4.7	Claude Fable	Use
ESPP, qualifying sale	2 of 21,556 tokens	2 of 2220 tokens	2 of 28,568 tokens	2 of 21,403 tokens	GPT-6 AstraTied on right answers, fewest tokens
Qualified dividends	2 of 22,366 tokens	2 of 2209 tokens	2 of 211,898 tokens	2 of 21,681 tokens	GPT-6 AstraTied on right answers, fewest tokens
Section 1256 contracts	4 of 42,465 tokens	4 of 4315 tokens	4 of 47,401 tokens	4 of 41,086 tokens	GPT-6 AstraTied on right answers, fewest tokens
Wash sales	4 of 41,681 tokens	4 of 4259 tokens	4 of 45,087 tokens	4 of 41,380 tokens	GPT-6 AstraTied on right answers, fewest tokens
ESPP, disposition	4 of 42,888 tokens	4 of 4712 tokens	4 of 417,002 tokens	4 of 42,621 tokens	GPT-6 AstraTied on right answers, fewest tokens
ISO and AMT	4 of 46,220 tokens	4 of 41,645 tokens	4 of 484,300 tokens	3 of 45,168 tokens	GPT-6 AstraTied on right answers, fewest tokens
Loss carryovers	4 of 41,606 tokens	4 of 4283 tokens	4 of 436,000 tokens	4 of 41,978 tokens	GPT-6 AstraTied on right answers, fewest tokens
Whole investor return	11 of 1231,781 tokens	9 of 126,048 tokens	12 of 12195,062 tokens	12 of 1227,796 tokens	Claude FableTied on right answers, fewest tokens
ISO crossover	2 of 25,607 tokens	1 of 21,445 tokens	2 of 251,451 tokens	2 of 25,050 tokens	Claude FableTied on right answers, fewest tokens
Lot selection	2 of 26,858 tokens	2 of 2735 tokens	2 of 228,138 tokens	2 of 24,969 tokens	GPT-6 AstraTied on right answers, fewest tokens

Data from our research: runs on 31 Oct and 4 Nov 2026, with 40 stock-market tax answers, each model at high effort, 2 to 12 per row. Claude Opus was not in these runs. Ranked by right answers, then by fewest reply tokens, so GPT-6 Astra takes 8 of 10 rows on ties, with the fewest right answers overall, 36 of 40. Rows this small show the form of the report. A project has more questions per row, and the tests say which gaps hold.

Area	Sub-task	Use
Biotech	Primer design	Model A
Biotech	Assay readout	Model B
Biotech	Pathway reasoning	Model A
Healthcare	Dosing calculations	Model C
Healthcare	Clinical coding	Model B
Healthcare	Trial eligibility	Model A

An illustration of the layout. We have not run this area, so this shows no results and no model names.

Area	Sub-task	Use
Chemistry	Stoichiometry	Model B
Chemistry	Reaction yield	Model A
Chemistry	Spectra reading	Model C
Pharma	Half-life and clearance	Model A
Pharma	Drug interactions	Model C
Pharma	Label checks	Model B

An illustration of the layout. We have not run this area, so this shows no results and no model names.

Area	Sub-task	Use
Law	Jurisdiction	Model A
Law	Case citations	Model C
Law	Statute deadlines	Model B
Contracts	Liability caps	Model B
Contracts	Termination rights	Model A
Contracts	Renewal dates	Model B

An illustration of the layout. We have not run this area, so this shows no results and no model names.

Area	Sub-task	Use
Manufacturing	Takt time	Model B
Manufacturing	Control charts	Model A
Manufacturing	Bill of materials	Model C
Transportation	Route plans	Model A
Transportation	Load limits	Model B
Transportation	Hours of service	Model C

An illustration of the layout. We have not run this area, so this shows no results and no model names.

Agents and tool calling

Each step can go to a different model.

An agent works in steps: it reads the request, plans, calls a tool, reads what comes back, and writes the answer. Each step can go to a different model. We test each step, and the report says which model to use for which one.

Step	What the report shows	Use
Read the request	Task understood, tokens, time	Model A
Plan the steps	A plan that works, tokens, time	Model B
Call a tool	Right tool, valid arguments, tokens, time	Model C
Read the result	Right facts taken from it, tokens, time	Model B
Write the answer	Right answer, tokens, time	Model A

An illustration of the layout, with no results and no model names.

Two real examples

What the answer looks like.

It is not just token costs that matter. It also depends on how efficient the models are with their tokens.

Example 1

Which frontier model for hard finance questions?

20 stock-market tax questions at four levels of difficulty, each asked with the cited law in the prompt and without it, so 40 in all. Claude Sonnet, Claude Fable, GPT-6 Astra and Grok 4.7, each at high effort. Claude Opus was not in these runs.

Right answers, of 40

Grok 4.7

Claude Sonnet

Claude Fable

GPT-6 Astra

Reply tokens, all 40 questions

Grok 4.7

444,907

Claude Sonnet

63,028

Claude Fable

53,132

GPT-6 Astra

11,871

What we would recommend. Start with the model that uses the fewest tokens, after checking each vendor's price per token, and run a larger question set before settling. The widest gap in right answers is Grok over GPT-6 Astra, where 4 questions went one way and none the other, with an exact McNemar p of 0.125. So 40 questions do not separate these models. The tokens differ by about 37 times, and the median question took 8.6 seconds on GPT-6 Astra against 124.7 on Grok.

Run through each vendor's own command-line tool, with X search on for Grok, on 31 Oct 2026, and on 4 Nov 2026 for Claude Fable. Five of the 40 prompts gained cited rules between the two dates. Tokens are as each tool reported them.

Example 2

Which effort level?

The same model, at different effort levels, on the same long calculations.

Claude Opus 5.5, two 300-step calculations: right answers

High effort

2 of 2

Max effort

0 of 2

Output tokens, both calculations

High effort

69,798

Max effort

654,684

What we would recommend. Use high effort for long calculations like these. At max, Claude Opus 5.5 reasoned past the 128,000 tokens one reply can hold and gave no answer. Do not assume the same on every task: on one 50-step calculation, medium, high and max were all right, with 7,584, 8,932 and 24,875 tokens, and GPT-5.5 went from 7 to 12 of 14 level-10 questions right from low to high effort.

Observed in Oct 2026, and published on our frontier models page.

Watch them answer

Real questions, real answers.

We picked these to show each case. They are saved answers, played back evenly across the time each model took.

Speed Recorded 31 Oct and 4 Nov 2026

Welcome. Let's get you started.

Frontier models

Frontier models, enhanced

Open models, your endpoint

Open models, your server

You choose a size and an area. We do the rest.

Place an order.

Build the questions.

Choose your areas

Let us look at your product

Describe your tasks

Send your questions or traces

The format: a CSV file

Sign off, or let our systems decide.

We run them.

Get the report.

Which model for which sub-task

Each step can go to a different model.

What the answer looks like.

Which frontier model for hard finance questions?

Right answers, of 40

Reply tokens, all 40 questions

Which effort level?

Claude Opus 5.5, two 300-step calculations: right answers

Output tokens, both calculations

Real questions, real answers.

Build a better AI system.

Start a conversation.