Decision Models
Compare 215 decision models from 118 organisations by accuracy, calibration, latency and cost per 1,000 decisions. 0 have an independent benchmark score; sort, filter, and open any model for details.
- Models
- 215
- Independently scored
- 0
- Hosted APIs
- 5
- New this month
- 128
- Cheapest hosted
- —
0 models · 215 without an independent score hidden (turn off “Scored only” to list them)
Decision Index = Jev Decision Index Full score (third-party, the headline). JevBench Capability = Benchmark Heaven's mean of Intelligence and Calibration (third-party; it also ranks hosted APIs the Index has not run). Latency is the Index's single-request median; hosted APIs include the network round trip. Cost assumes 200 input tokens per decision. “Scored only” hides models with no independent benchmark score; every model keeps its own page.
Frequently asked questions
What is a decision model?
A decision model (also called a System One model) takes a piece of text and a set of typed questions — pick one option, place it on a scale, or answer yes or no — and returns a probability for every allowed answer instead of generating text. TypeSafe’s Jev defined the category in September 2026.
How many decision models are there?
We list 215 from 118 organisations: 5 available only as a hosted API, 207 as open weights only, and 0 both. 128 were released this month. 0 have at least one independent benchmark score; the other 215 are listed but hidden from the leaderboard until a board measures them.
Which decision model is the most accurate?
On the Jev Decision Index, which scores 0 of the models listed here, the top-ranked model is shown in the table. On Benchmark Heaven JevBench, which also measures hosted APIs (0 models), the highest Capability is shown in the table. Check calibration (ECE) and latency alongside the score.
Which benchmarks do you use?
The headline is the Jev Decision Index: the largest third-party board, with private held-out sets weighted 80% and calibration reported. The second column is Benchmark Heaven JevBench Capability (the mean of its Intelligence and Calibration axes), which covers the hosted APIs the Index has not run. DecisionBench, Enterprise Decision Arena and the JevBench coherence test are shown on model pages as cross-checks. Scores a publisher reports about its own model are labelled self-reported and never ranked.
What is the cheapest hosted decision model?
Among hosted APIs, the lowest cost per 1,000 decisions listed here is shown in the table, assuming 200 input tokens per decision. Open-weights models cost only the compute you run them on.
What does ECE mean?
Expected calibration error: the average gap between how confident a model says it is and how often it is right. Lower is better; below about 0.05, a stated 80% confidence is right about 80% of the time.
How often is this data updated?
Daily. Hosted models and prices come from OpenRouter and the Vercel AI Gateway; scores from the Jev Decision Index, Benchmark Heaven, DecisionBench and Enterprise Decision Arena; new models found on Hugging Face are reviewed before they are listed.