A 149.6M ModernBERT-base cross-encoder with a scalar head that picks an agent's next action or tool from labels and descriptions you supply. Each option is scored against the state and a softmax ranks the set; no text is generated. Experimental preview, not calibrated.
Nandakishor M's typed-decision model. A frozen Qwen3.5 backbone is read, never trained, and cut off mid-stack; a small trained physics engine settles each choice, score or yes/no question into calibrated probabilities, a conformal set and an abstain flag. 0.8B and 4B.
Decides
choice, rank, route
choice, score, noul, classify, route
Architecture
modernjev
vega
Fine-tuned from
answerdotai/modernbert-base
qwen/qwen3.5-0.8b
License
apache-2.0
apache-2.0
Availability
Open weights
Open weights
Hosted by
—
—
Input price
—
—
Decision accuracy
73.4%
55.6%
Calibration error
—
0.151
Valid action rate
—
—
Median latency
—
—
p95 latency
—
Figures are from each model’s manifest; accuracy and latency are what the publishers report, on their own suites and hardware. Add a third model.
Questions
What is the difference between modernjev-decide and vega?
modernjev-decide is from Maziyar Panahi and vega from Nandakishor M. Both have open weights you can download and run. Both answer choice and route questions. Only modernjev-decide answers rank. Only vega answers score, noul and classify. vega reads up to 72K tokens of state, against 4K tokens for modernjev-decide. modernjev-decide is the smaller model, at 150M parameters to 800M.
Which is more accurate, modernjev-decide or vega?
They report on different suites — modernjev-decide 73.4% on AgentToolDecisions-180K held-out test, next action type (1,158 decisions, 3 choices; the maker's split), vega 55.6% on His 11 October 2026 benchmark, published with its results and its harness (a notebook in vegaml): macro accuracy over 43 public tasks for the 0.8B, 0.717 for the 4B; with JevBench (231 items) and DecisionBench (1,075 rows), 3,841 decisions in all, seed 20261011. On his own set of 2,050 choice decisions over six intent workflows (esci, clinc150, appliances, banking77, when2call, apibank), one choice question each, which he has not published with its question wording, he reports 0.763 (calibration error 0.026) for the 0.8B with its public_intents adapter, 0.389 without the adapter and 0.803 for the 4B with it. He publishes no accuracy for the typed_decisions adapters — so the numbers do not rank them. Test both on your own labelled examples.
Which is cheaper, modernjev-decide or vega?
modernjev-decide: Free (open weights). vega: Free (open weights). Open weights cost nothing per call beyond your own hardware.
Can I run modernjev-decide or vega locally?
Yes, both: systemone pull maziyar-panahi/modernjev-decide and systemone pull nandakishorm/vega download the weights.
—
Evaluation suite
AgentToolDecisions-180K held-out test, next action type (1,158 decisions, 3 choices; the maker's split)
His 11 October 2026 benchmark, published with its results and its harness (a notebook in vegaml): macro accuracy over 43 public tasks for the 0.8B, 0.717 for the 4B; with JevBench (231 items) and DecisionBench (1,075 rows), 3,841 decisions in all, seed 20261011. On his own set of 2,050 choice decisions over six intent workflows (esci, clinc150, appliances, banking77, when2call, apibank), one choice question each, which he has not published with its question wording, he reports 0.763 (calibration error 0.026) for the 0.8B with its public_intents adapter, 0.389 without the adapter and 0.803 for the 4B with it. He publishes no accuracy for the typed_decisions adapters
