Decision Leaderboard.
Pick a job. Then see which decision models do it best, how fast, and at what cost.
Flagging harmful requests, harmful replies and refusals. This is the job behind content moderation and guardrails. Most models score high here, so speed and cost often decide.
Test: WildGuardTest by Ai2 · Run by Artificial Analysis · Source ↗
d1-ultraLiquid AI95#189 ms$22
Decider V1.2Perplexity94#4205 ms$7.1
Jev 1.13TypeSafe94#4125 ms$23
Elephant 1Baseten93#7151 ms$6.3- Decider Flash V1.2Perplexity93#7142 ms$6.0
- Jev 1.14TypeSafe93#776 ms$23
ClefCloudflare91#11279 ms$107- Clef FlashCloudflare91#11247 ms$17
- d1Liquid AI91#11298 ms$10
Mercury DecideInception88#14454 ms$6.3
Microsoft-Decision-1Microsoft88#14204 ms$12- Clef OmniCloudflare87#16163 ms$66
GPT-6 Luna DecisionsOpenAI83#17159 ms$48
- d1-ultra#1
- F1 score
- 95
- Speed
- 89 ms
- Cost / 1M
- $22
- Decider V1.2#4
- F1 score
- 94
- Speed
- 205 ms
- Cost / 1M
- $7.1
- Jev 1.13#4
- F1 score
- 94
- Speed
- 125 ms
- Cost / 1M
- $23
- Elephant 1#7
- F1 score
- 93
- Speed
- 151 ms
- Cost / 1M
- $6.3
- Decider Flash V1.2#7
- F1 score
- 93
- Speed
- 142 ms
- Cost / 1M
- $6.0
- Jev 1.14#7
- F1 score
- 93
- Speed
- 76 ms
- Cost / 1M
- $23
- Clef#11
- F1 score
- 91
- Speed
- 279 ms
- Cost / 1M
- $107
- Clef Flash#11
- F1 score
- 91
- Speed
- 247 ms
- Cost / 1M
- $17
- d1#11
- F1 score
- 91
- Speed
- 298 ms
- Cost / 1M
- $10
- Mercury Decide#14
- F1 score
- 88
- Speed
- 454 ms
- Cost / 1M
- $6.3
- Microsoft-Decision-1#14
- F1 score
- 88
- Speed
- 204 ms
- Cost / 1M
- $12
- Clef Omni#16
- F1 score
- 87
- Speed
- 163 ms
- Cost / 1M
- $66
- GPT-6 Luna Decisions#17
- F1 score
- 83
- Speed
- 159 ms
- Cost / 1M
- $48
Ranks are out of 17 models Artificial Analysis tests, some not on Recurrence · Not tested · Mammoth 1, Kev 4B, Drex Max, Drex v2.0, Drex v1.5, Span-02, Span-01, Span-01 Lite, Classify-1, Tev1 4B Experimental, Solar Decide, Solar Decide Flash
No public test yet.
These are common jobs for decision models, but no established benchmark covers them. Until one does, compare models on your own examples in the Decision Arena.
Ticket routing
Which team should handle this ticket?
LLM as judge
How good is this agent’s reply?
Search reranking
Which document is the most relevant?
Sentiment
What is the overall sentiment?
Lead scoring
How qualified is this lead?
Where the numbers come from.
We copy published results and don't run these tests ourselves. Scores from different tests aren't comparable, so compare models within one scenario.
Artificial Analysis Decision Index
Independent tests of decision models on four public and private evaluations, with the median time and average cost Artificial Analysis measured for each.
As of Oct 24, 2026 · See source terms · Source ↗
OpenRouter Decision Model Rankings
Weekly decision requests on OpenRouter. Shows what people use, not how good a model is. Only the top 10 are published.
Source: OpenRouter (openrouter.ai/rankings), as of Oct 24, 2026. Licensed under CC BY 4.0.