Task ranking
Best Decision Model for Agent Routing and Scoring
Sage 1.3.0 (Levanto Labs) is the provisional first pick for Best Decision Model for Agent Routing and Scoring as of October 16, 2026, while independent task measurements are still limited.[1][2] Sage 1.3.0 leads the JevBench composite ranking with the highest overall score of 74.05, combining strong calibration (92) with moderate decision accuracy (66). H2O-Lightning-4B v1.1 (H2O.ai, Qwen3.5-4B fine-tune, stock vLLM + open shim) is the next-ranked alternative. It scores 72.52 on JevBench (#2 of 114), Benchmark Heaven's composite of decision accuracy, calibration, speed, and cost for typed decision models; its decision accuracy axis is 60 and its calibration axis 90, each out of 100.
About this recommendation
- Updated
- Evidence through
- Sources
- 8
- Revision
- v17
Comparing Jev alternatives or choosing a System One model? Read our guide to AI decision models, routing, classification and scoring for the decision model vs LLM tradeoffs and a self-hosting checklist.
Decision audit
Why this result
Inspect the inputs and the computed order behind the recommendation.
Models screened
24
live candidates
Evaluation feeds
1
task-weighted
Winner coverage
100%
intended feed weight
Largest provider share
1 of 8
chinhnc
Sage 1.3.0 (Levanto Labs) leads the measured task comparison with limited independent task measurements. Community volume, release age and usage popularity add no ranking points.
Real-world track record is still developing. The recommendation can lead on independent measurements while practitioner reports accumulate.
Sources evaluated
These are the weights used in scoring. Missing measurements are omitted from each model’s average. A feed with no result for any model released in the last 60 days is stale and does not score until it catches up.
| Evaluation feed | Weight | Winner result | Field measured |
|---|---|---|---|
| JevBenchObserved 2026-10-16 · evaluation date not supplied · Sage 1.3.0 (Levanto Labs) · standard error 0.88 | 100% | #1 | 24/24 |
Provider concentration
Each exact model is scored separately; provider identity is not a ranking input.
- chinhnc1 model
- eldanring1 model
- h2oai1 model
- levanto-labs1 model
- mapika1 model
- torchcast-ai1 model
- typesafe-ai1 model
- wity1 model
Decision table
Every published model is shown in computed order. Practitioner sources are distinct community threads, not the citations repeated in the prose below.
levanto-labs · coverage 100% · provisional evidence
#1 JevBench
h2oai · coverage 100% · provisional evidence
#2 JevBench
typesafe-ai · coverage 100% · provisional evidence
#3 JevBench
chinhnc · coverage 100% · provisional evidence
#4 JevBench
wity · coverage 100% · provisional evidence
#5 JevBench
mapika · coverage 100% · provisional evidence
#6 JevBench
torchcast-ai · coverage 100% · provisional evidence
#7 JevBench
eldanring · coverage 100% · provisional evidence
#8 JevBench
Task score averages the available, task-weighted measurements. Benchmark positions compare models matched to our catalog. Models measured by at least two independent evaluator families are considered first. Missing tests, practitioner report counts, popularity and release age add no points. Coverage describes evidence breadth, not a probability of correctness. Scores within two points are shown as close under this policy; this is not a statistical confidence interval. When the leading scores are that close, the model measured by more evaluator families ranks first.
- 01
Sage 1.3.0 (Levanto Labs)levanto-labs
Sage 1.3.0 leads the JevBench composite ranking with the highest overall score of 74.05, combining strong calibration (92) with moderate decision accuracy (66).
Best when: Use for agent routing where calibrated confidence matters, as it scores 92 on JevBench's calibration axis.No practitioner reports are linked yet, so no cautions are listed. Absence of reports is not evidence of reliability.
H2O-Lightning-4B v1.1 (H2O.ai, Qwen3.5-4B fine-tune, stock vLLM + open shim) scores 72.52 on JevBench (#2 of 114), Benchmark Heaven's composite of decision accuracy, calibration, speed, and cost for typed decision models; its decision accuracy axis is 60 and its calibration axis 90, each out of 100.
Best when: Its strongest result for this task is on JevBench, which measures calibrated typed decisions.No practitioner reports are linked yet, so no cautions are listed. Absence of reports is not evidence of reliability.
- 03
Jev 1.13.0 (TypeSafe AI)typesafe-ai
Jev 1.13.0 places third on JevBench with 71.49 composite, and independent evaluation shows 0.028 pooled ECE across 22 choice datasets with substantial variance (0.003 to 0.279) and under-confidence on yes/no tasks.
Best when: Filter to its most confident half of predictions to boost accuracy dramatically, for example from 79.7% to 96.3% on Banking77.Watch out for
- Watch for variable calibration across datasets, with ECE ranging from 0.003 to 0.279, and slight under-confidence on binary yes/no decisions.
Quyet-1.0-Large (Chinh Nguyen, Gemma-4-31B decoder, option-letter logits) scores 71.39 on JevBench (#4 of 114), Benchmark Heaven's composite of decision accuracy, calibration, speed, and cost for typed decision models; its decision accuracy axis is 73 and its calibration axis 90, each out of 100.
Best when: Its strongest result for this task is on JevBench, which measures calibrated typed decisions.No practitioner reports are linked yet, so no cautions are listed. Absence of reports is not evidence of reliability.
wity-1 (Wity, reasoning auto) scores 70.77 on JevBench (#5 of 114), Benchmark Heaven's composite of decision accuracy, calibration, speed, and cost for typed decision models; its decision accuracy axis is 70 and its calibration axis 88, each out of 100.
Best when: Its strongest result for this task is on JevBench, which measures calibrated typed decisions.No practitioner reports are linked yet, so no cautions are listed. Absence of reports is not evidence of reliability.
- 06
decider-12b v2 (Mapika)mapika
decider-12b v2 (Mapika) scores 70.42 on JevBench (#6 of 114), Benchmark Heaven's composite of decision accuracy, calibration, speed, and cost for typed decision models; its decision accuracy axis is 61 and its calibration axis 84, each out of 100.
Best when: Its strongest result for this task is on JevBench, which measures calibrated typed decisions.No practitioner reports are linked yet, so no cautions are listed. Absence of reports is not evidence of reliability.
- 07
torchcast-decision-12b (Torchcast AI, Gemma-4-12B fine-tune, option-letter logprob readout)torchcast-ai
torchcast-decision-12b (Torchcast AI, Gemma-4-12B fine-tune, option-letter logprob readout) scores 69.91 on JevBench (#7 of 114), Benchmark Heaven's composite of decision accuracy, calibration, speed, and cost for typed decision models; its decision accuracy axis is 61 and its calibration axis 83, each out of 100.
Best when: Its strongest result for this task is on JevBench, which measures calibrated typed decisions.No practitioner reports are linked yet, so no cautions are listed. Absence of reports is not evidence of reliability.
- 08
Winnow-12B Q8eldanring · open-weight
Winnow-12B Q8 scores 68.89 on JevBench (#8 of 114), Benchmark Heaven's composite of decision accuracy, calibration, speed, and cost for typed decision models; its decision accuracy axis is 60 and its calibration axis 83, each out of 100.
Best when: Its strongest result for this task is on JevBench, which measures calibrated typed decisions.No practitioner reports are linked yet, so no cautions are listed. Absence of reports is not evidence of reliability.
Frequently asked
- What is the best model for Best Decision Model for Agent Routing and Scoring?
- Sage 1.3.0 (Levanto Labs) ranks first as of October 16, 2026. Sage 1.3.0 leads the JevBench composite ranking with the highest overall score of 74.05, combining strong calibration (92) with moderate decision accuracy (66).[1]
- What is the best alternative to Sage 1.3.0 (Levanto Labs)?
- H2O-Lightning-4B v1.1 (H2O.ai, Qwen3.5-4B fine-tune, stock vLLM + open shim) ranks second. H2O-Lightning-4B v1.1 (H2O.ai, Qwen3.5-4B fine-tune, stock vLLM + open shim) scores 72.52 on JevBench (#2 of 114), Benchmark Heaven's composite of decision accuracy, calibration, speed, and cost for typed decision models; its decision accuracy axis is 60 and its calibration axis 90, each out of 100.[2]
- Which open-weight model ranks highest for Best Decision Model for Agent Routing and Scoring?
- Winnow-12B Q8 is the highest-ranked open-weight model, at #8. Winnow-12B Q8 scores 68.89 on JevBench (#8 of 114), Benchmark Heaven's composite of decision accuracy, calibration, speed, and cost for typed decision models; its decision accuracy axis is 60 and its calibration axis 83, each out of 100.[3]
Sources
- 1
“Within the 114 catalog models matched to this evaluation: Scores 74.05 on JevBench (#1 of 114), Benchmark Heaven's composite of decision accuracy, calibration, speed, and cost for typed decision models; its decision accuracy axis is 66 and its calibration axis 92, each out of 100.”
JevBench · Benchmark · Oct 16, 2026 - 2
“Within the 114 catalog models matched to this evaluation: Scores 72.52 on JevBench (#2 of 114), Benchmark Heaven's composite of decision accuracy, calibration, speed, and cost for typed decision models; its decision accuracy axis is 60 and its calibration axis 90, each out of 100.”
JevBench · Benchmark · Oct 16, 2026 - 3
“Within the 114 catalog models matched to this evaluation: Scores 68.89 on JevBench (#8 of 114), Benchmark Heaven's composite of decision accuracy, calibration, speed, and cost for typed decision models; its decision accuracy axis is 60 and its calibration axis 83, each out of 100.”
JevBench · Benchmark · Oct 16, 2026 - 4
“Two outside sources on Jev, and what they mean for this measurement. Neither is TypeSafe material. - An evaluation paper (Deußer, Sparrenberg and Sifa, arXiv 2609.37647) tested jev-1.13.0 on 37 datasets. Pooled calibration error (ECE) over 22 choice datasets is 0.028, but it is from 0.003 to 0.279 for single datasets. Yes/no answers are slightly under-confident. When Jev answers only its most confident half, accuracy rises, for example Banking77 from 79.7% to 96.3%. - A third-party explainer…”
mohanraj00 · GitHub · Oct 4, 2026 - 5
“Within the 114 catalog models matched to this evaluation: Scores 71.39 on JevBench (#4 of 114), Benchmark Heaven's composite of decision accuracy, calibration, speed, and cost for typed decision models; its decision accuracy axis is 73 and its calibration axis 90, each out of 100.”
JevBench · Benchmark · Oct 16, 2026 - 6
“Within the 114 catalog models matched to this evaluation: Scores 70.77 on JevBench (#5 of 114), Benchmark Heaven's composite of decision accuracy, calibration, speed, and cost for typed decision models; its decision accuracy axis is 70 and its calibration axis 88, each out of 100.”
JevBench · Benchmark · Oct 16, 2026 - 7
“Within the 114 catalog models matched to this evaluation: Scores 70.42 on JevBench (#6 of 114), Benchmark Heaven's composite of decision accuracy, calibration, speed, and cost for typed decision models; its decision accuracy axis is 61 and its calibration axis 84, each out of 100.”
JevBench · Benchmark · Oct 16, 2026 - 8
“Within the 114 catalog models matched to this evaluation: Scores 69.91 on JevBench (#7 of 114), Benchmark Heaven's composite of decision accuracy, calibration, speed, and cost for typed decision models; its decision accuracy axis is 61 and its calibration axis 83, each out of 100.”
JevBench · Benchmark · Oct 16, 2026
Rankings synthesized from community evidence and open benchmarks. See methodology. Not driven by vendor marketing.