A 0.4B decision model made from the first 20 layers of Qwen3-0.6B plus a small attention head that compares options. It answers choice, noul and score questions in one pass, and choice order cannot change its answer by construction. Non-commercial licence.
Telnyx's hosted decision models, in beta on Telnyx Inference. One request carries shared state and up to 64 typed questions (choice, yes/no noul, ordered score) and returns a probability per option, with no generated text, on a TypeSafe-compatible /v1/systemone route.
Decides
choice, score, noul
choice, score, noul, classify, route
Architecture
bev-decider
telnyx-decision
Fine-tuned from
qwen/qwen3-0.6b
—
License
cc-by-nc-4.0
proprietary
Availability
Open weights
Hosted API
Hosted by
—
Telnyx
Input price
—
$0.035/MTok
Decision accuracy
74.7%
—
Calibration error
—
—
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 bev-decider and decision-flash?
bev-decider is from Avishek Biswas and decision-flash from Telnyx. bev-decider has open weights you can download and run; decision-flash is only available as a hosted API. Both answer choice, score and noul questions. Only decision-flash answers classify and route. bev-decider is licensed cc-by-nc-4.0; decision-flash, proprietary.
Which is more accurate, bev-decider or decision-flash?
Only bev-decider publishes an accuracy figure (74.7% on avbiswas/bev-decision test split (5,000 held-out questions over 2,617 states), the maker's own); decision-flash does not, so there is no comparison to make without your own test.
Which is cheaper, bev-decider or decision-flash?
bev-decider: Free (open weights). decision-flash: $0.035 / $0 per 1M. Open weights cost nothing per call beyond your own hardware.
Can I run bev-decider or decision-flash locally?
bev-decider yes — systemone pull avishek-biswas/bev-decider downloads its weights. The other is only served as a hosted API.
—
Evaluation suite
avbiswas/bev-decision test split (5,000 held-out questions over 2,617 states), the maker's own
