Start building ↗

JevBench leaderboard

#1 on JevBench.

  1. #1Sage 1.3.074.0
  2. #4Jev 1.13.071.5
  3. #6Microsoft-Decision-1Azure Foundry69.1
  4. #9OpenAI Decisions62.5

Composite score, 0–100. Source: JevBench Decision API leaderboard (Benchmark Heaven), board v1.7.46, 14 Oct 2026

Sage 1.4 vs Jev

Leading key benchmarks.

Decision panel

11 benchmarks, 7,210 rows

Sage 1.489.4%
Jev84.5%

JevBench public

213 tasks

Sage 1.4203 / 213
Jev186 / 213

Multistep decisions

100 tasks

Sage 1.494 / 100
Jev71 / 100

Decision Index 0.2.1

reasoning off

Sage 1.461.0
Jev57.9

Sage 1.4 with reasoning auto (Decision Index: off). Jev: published scores; JevBench public and multistep decisions are our runs of Jev 1.13.0.

Decision panel

11 benchmarks, 7,210 rows

BenchmarkRowsJevSage 1.4
reasoning off		Sage 1.4
reasoning auto
WinoGrande1,00090.788.287.4
FinancialPhraseBank99977.096.897.1
RAGTruth1,50077.387.487.7
JudgeBench35078.679.185.1
BBH75094.385.391.2
JevBench public hard10173.378.289.1
TabFact50089.891.491.8
ContractNLI51077.577.180.0
Circa50084.689.689.6
Belebele50095.096.896.6
TruthfulQA binary50092.082.883.2
Overall7,21084.588.189.4

Accuracy, %. Best per row in bold. Jev: published scores.

Sage Router

Same accuracy. A third of the cost.

84%86%88%90%92%94%96%$0.001$0.01$0.1Cost per question (log scale)GPQA Diamond accuracyJev Router: 93.9% at $0.016Jev RouterGPT-6.1 Sol (single model): 93.4% at $0.007GPT-6.1 SolPareto: 92.4% at $0.011ParetoFugu Max: 88.6% at $0.025Fugu MaxOpenRouter Auto Low: 88.2% at $0.009OpenRouter Auto LowOpenRouter Auto Medium: 87.6% at $0.011OpenRouter Auto MediumOpenRouter Auto High: 83.2% at $0.061OpenRouter Auto HighFugu Ultra · 94.6% at $0.38Fugu Ultra: 94.6% at $0.38Claude Opus 5.5 alone (no router): 89.4% at $0.018Opus 5.5 aloneSage Router, CQT 0: 91.6% at $0.0157Sage Router, CQT 1: 92% at $0.01471Sage Router, CQT 2: 93.5% at $0.01112Sage Router, CQT 3: 93.5% at $0.00523Sage Router, CQT 4: 89% at $0.00234Sage Router, CQT 5: 87.5% at $0.0015Sage Router, CQT 6: 87.5% at $0.0007Sage Router, CQT 8: 87.5% at $0.0006Sage Router, CQT 10: 87.4% at $0.0006Sage RouterCQT 0 · best qualityCQT 10 · lowest cost84%88%92%96%$0.001$0.01$0.1$ per question (log)GPQA Diamond accuracyJev Router: 93.9% at $0.016JevGPT-6.1 Sol (single model): 93.4% at $0.007GPT-6.1 SolPareto: 92.4% at $0.011ParetoFugu Max: 88.6% at $0.025Fugu MaxOpenRouter Auto Low: 88.2% at $0.009Auto LowOpenRouter Auto Medium: 87.6% at $0.011Auto MedOpenRouter Auto High: 83.2% at $0.061Auto HighFugu Ultra · $0.38Fugu Ultra: 94.6% at $0.38Claude Opus 5.5 alone (no router): 89.4% at $0.018Opus aloneSage Router, CQT 0: 91.6% at $0.0157Sage Router, CQT 1: 92% at $0.0147Sage Router, CQT 2: 93.5% at $0.0111Sage Router, CQT 3: 93.5% at $0.0052CQT 3Sage Router, CQT 4: 89% at $0.0023Sage Router, CQT 5: 87.5% at $0.001Sage Router, CQT 6: 87.5% at $0.0007Sage Router, CQT 8: 87.5% at $0.0006Sage Router, CQT 10: 87.4% at $0.0006Sage Routerbest qualitylowest cost
GPQA DiamondAccuracy$ / questionCost vs Opus alone
Claude Opus 5.5 alone · no router89.4%$0.018baseline
Sage Router · CQT 393.5%$0.005−72%
Sage Router · CQT 293.5%$0.011−39%
Jev Router93.9%$0.016−11%
Pareto92.4%$0.011−39%
OpenRouter Auto88.2%$0.009−50%
Sage Router · CQT 887.5%$0.0006−97%

Source: OpenRouter router leaderboard, 4 Oct 2026. GPQA Diamond, 990 tasks; costs at OpenRouter prices; differences under ~4 points are within the margin of error. Claude Opus 5.5 alone is the same board's single-model baseline (reasoning high, no router), read 19 Oct 2026.

Levanto Guardrail

GPT-5 safety. Five times faster.

868890929496981005060708090% of harmless requests allowed → higher is better% of harmful prevented → higher is better← idealLevanto Sage, 77% prevented, 95.5% allowed, 208msLevanto Sage208msOpenAI GPT-5, 76.3% prevented, 96% allowed, 1009msOpenAI GPT-51009msGoogle Model Armor, 89.7% prevented, 87.5% allowed, 195msGoogle Model Armor195msOpenAI GPT-5-mini, 56.1% prevented, 100% allowed, 1114msOpenAI GPT-5-mini1114msMeta PromptGuard, 57.3% prevented, 95.5% allowed, 301msMeta PromptGuard301msInjecGuard, 55.8% prevented, 95.5% allowed, 262msInjecGuard262msProtect AI, 54.7% prevented, 95.5% allowed, 312msProtect AI312msAWS Bedrock LOW, 58.8% prevented, 88.1% allowed, 191msAWS Bedrock LOW191msAWS Bedrock MED, 51.6% prevented, 96% allowed, 188msAWS Bedrock MED188msUnguarded GPT-4o, 51.1% prevented, 100% allowedUnguarded GPT-4o
77%
Harm prevented
95.5%
Harmless allowed
208 ms
Median latency
~5×
Faster than GPT-5, similar quality

AgentHarm: harmful agent tasks to stop, harmless ones to let through. Full benchmark ↗

An engraved blue sea and mountain landscape under a rising sun.

Intelligence, at runtime

Sage / Levanto

Start free.

Sign in or create an account.

Continue with Google
or with email

We’ll email you a 6-digit code.