Shanghai AI Laboratory's multimodal decision model. A fine-tune of Qwen3.5-4B's language backbone (vision tower frozen) that answers up to 16 Choice, Score and Noul questions about a state and up to eight images in one forward pass, with a fitted calibration temperature.
Small open typed-decision model with calibrated probabilities, first-class Nepali coverage.
Decides
choice, score, noul, classify, route
choice, score, noul, classify, route
Architecture
intern-decision
mira
Fine-tuned from
qwen/qwen3.5-4b
jhu-clsp/mmbert-small
License
apache-2.0
apache-2.0
Availability
Open weights
Open weights
Hosted by
—
—
Input price
—
—
Decision accuracy
80.6%
74.5%
Calibration error
—
0.025
Valid action rate
—
100.0%
Median latency
44 ms
30 ms
p95 latency
44.6 ms
43 ms
Evaluation suite
LocalLLaMA/typed-decisions test split (400 cases, 2,000 decisions)
s1-decision-bench
Latest version
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 intern-decision and mira?
intern-decision is from InternLM (Shanghai AI Laboratory) and mira from SAGEA. Both have open weights you can download and run. Both answer choice, score, noul, classify and route questions.
Which is more accurate, intern-decision or mira?
They report on different suites — intern-decision 80.6% on LocalLLaMA/typed-decisions test split (400 cases, 2,000 decisions), mira 74.5% on s1-decision-bench — so the numbers do not rank them. Test both on your own labelled examples.
Which is cheaper, intern-decision or mira?
intern-decision: Free (open weights). mira: Free (open weights). Open weights cost nothing per call beyond your own hardware.
Which is faster, intern-decision or mira?
By their publishers’ figures, mira answers in about 30 ms at the median and intern-decision in about 44 ms — measured on different hardware, so treat it as a rough guide.
Can I run intern-decision or mira locally?
Yes, both: systemone pull internlm/intern-decision and systemone pull sagea/mira download the weights.
