Pick a model

Which decision model fits your job?

The best model overall isn't always the best for you. Say what the decisions are about and what you can live with, and the 5 benchmarks turn into a shortlist.

What are the decisions about?

Adjust what matters

SudokuLogical reasoningSomeRulebookFollowing written rulesSomeConsistencyNot getting thrown offSomeOddsWeighing riskSomeFairnessTreating customers alikeSome

What can you live with?

Cost per 1,000 tickets, at most

Time per ticket, at most

Fairness score, at least

Follows instructions planted in a message, at most

General LLMs

A ticket is one Rulebook request: a one-page rulebook, a customer's ticket and three decisions. Fairness and planted instructions come from the Fairness and Consistency benchmarks.

Best fit

d1 UltraLiquid

90 / 100

Per 1,000 tickets: $0.0401
Per ticket: 0.28s
Fairness: 99.7%
Planted text followed: 0.0%

Sudoku84.0%
Rulebook98.6%
Consistency99.7%
Odds54.3%
Fairness99.7%

Every result for d1 Ultra →

Shortlist

Models that fit, best first
#	Model	Your score	Per 1k	Per ticket
1	d1 UltraLiquid	90	$0.0401	0.28s
2	Drex MaxNace.AI	84	$0.0643	0.40s
3	Decider v1.2Perplexity	75	$0.0191	0.37s
4	Jev 1.14TypeSafe	69	$0.0513	0.23s
5	ClefCloudflare	63	$0.0619	0.32s
6	Span-02Respan	62	$0.0254	1.15s
7	Mercury DecideInception	61	$0.0208	0.40s
8	Decision-1Microsoft	59	$0.0366	0.31s
9	GPT-6 Luna DecisionsOpenAI	58	$0.116	0.35s
10	Drex v2.0Nace.AI	58	$0.0517	0.41s

And 8 more that fit.

Ruled out: 4 models

DeepSeek V4.1 FlashGeneral LLMs left out
GPT-6 LunaGeneral LLMs left out
Claude Haiku 5.5General LLMs left out
Gemini 3.8 FlashGeneral LLMs left out

Share these choices:https://www.decisionbench.lol/pick

Scores use the ranking's scale: on each benchmark the best decision model is 100 and the worst 0, and a general LLM past either end counts as that end. Your weights decide how much each benchmark counts. Use this to make a shortlist, then test the top few on a few hundred of your own decisions before you trust one.