# ANT decision model by Semantique Team

> ANT adds trained decision heads to frozen Qwen backbones. Read its synthetic tests, validation-fitted escalation and separate backbone downloads.

Updated 2026-11-12 (Australia/Adelaide)

## Quick answers

### What is ANT by Semantique Team?

[ANT](<https://decisionmodels.ai/models/ant>): Trained candidate-scoring heads for frozen Qwen3.5-0.8B and 4B backbones.

[ANT facts](<https://decisionmodels.ai/models/ant>)

[Source](<https://huggingface.co/semantique-team/ant-4b>)

### Who makes ANT?

[ANT](<https://decisionmodels.ai/models/ant>): Semantique Team

[ANT facts](<https://decisionmodels.ai/models/ant>)

[Source](<https://huggingface.co/semantique-team/ant-4b>)

By Semantique Team · Hugging Face weight release: 2026-11-10 · OPEN

Trained candidate-scoring heads for frozen Qwen3.5-0.8B and 4B backbones. Each package contains the head, temperature calibration and an execute-or-escalate policy, with a separate backbone download. The English/Italian training and evaluation situations are synthetic.

## Facts

- Package: Head-only releases; [ANT-08B](<https://decisionmodels.ai/models/ant>) has 1,053,696 trainable parameters and a 4.2 MB head, excluding the backbone
- Inputs: Text state, question and supplied candidates; configured context budget 1,024 tokens and 128 per candidate
- Runtime: Python runtime from the matching [ant-08b](<https://decisionmodels.ai/models/ant>) or [ant-4b](<https://decisionmodels.ai/models/ant>) GitHub repository; optional MLX path
- Status: Work in progress; one training run per size, without measured seed variance
- Licence: Apache-2.0 for released heads; Qwen backbones retain their own Apache-2.0 licence

## Claims and sources

- **VENDOR-REPORTED**: The maker reports test accuracy 67.44% for [ANT-08B](<https://decisionmodels.ai/models/ant>) and 73.92% for [ANT-4B](<https://decisionmodels.ai/models/ant>) on 648 synthetic situations. On 456 situations from held-out use cases, accuracy is 57.02% and 59.65%. These are separate from the validation-fitted escalation policy. [source](<https://huggingface.co/semantique-team/ant-4b>)
- **VENDOR-REPORTED**: [ANT-08B](<https://decisionmodels.ai/models/ant>) reports validation coverage 44.2% at 9.9% error among accepted decisions, using 688 validation rows. Its fitted threshold does not establish that error rate on new domains; OOD ECE is 0.220. [source](<https://huggingface.co/semantique-team/ant-08b>)

## Links

- [Source: ANT-4B card](<https://huggingface.co/semantique-team/ant-4b>)
- [Source: ANT-08B card](<https://huggingface.co/semantique-team/ant-08b>)
- [Source: 4B metrics](<https://huggingface.co/semantique-team/ant-4b/blob/main/metrics.json>)
- [Source: 08B policy](<https://huggingface.co/semantique-team/ant-08b/blob/main/escalation.json>)
- [Source: Licence](<https://huggingface.co/semantique-team/ant-08b/blob/main/LICENSE>)

## Related projects

No related projects are listed in the registry.