Decision modelOpen weights
Strands Decider 2B
AWS Strands
- Released
- October 5, 2026
- Data date
- October 7, 2026
Strands Decider 2B scores supplied answer options in parallel. The v21 checkpoint adds a LoRA adapter and pointer head to Qwen3.5-2B-Base, returning scores for routing, tool selection, and other bounded decisions.
The checkpoint can take images through the base model’s vision tower, but the decision adapter was not trained on images. The Apache-2.0 release contains the adapter and decision head, not a standalone full language model.
Specifications and access
| Specification | Value and source |
|---|---|
| Model class | System One decision model with structured scores instead of free-form text generationSource |
| Input modalities | Text and images through the base model’s vision tower; the decision adapter was not trained on imagesSource |
| Output modalities | Structured Choice, Noul, and Score decisions with probabilities, no free-form response textSource |
| Checkpoint | StrandsAgents/strands-decider-2B-hobson-v21Source |
| Checkpoint release | v21 released October 5, 2026; the October 1 announcement described v19Source |
| Architecture | LoRA adapter and pointer decision head on Qwen/Qwen3.5-2B-Base; not a standalone full language-model checkpointSource |
| License | Apache-2.0Source |
Page 1 of 7
Measurements without matching peer values
These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.
BFCL · Decision Index 0.3 (Show measurement, test conditions, and source)
- Source value
- 0.8135
- Score
- 0.8135
- Metric
- case exact accuracy
- Unit
- ratio
- Benchmark version
- Decision Index release-v3
- Category
- decision-index-tools
- Direction
- Higher is better
- Sample
- 1694
- Source type
- independent-evaluator
- Evaluator
- Jev Decision Index 0.3 public-results data
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Independent evaluation. Jev Decision Index 0.3 release-v3 evaluation of exact model https://huggingface.co/StrandsAgents/strands-decider-2B-hobson-v21. The evaluator sampled one in six public-suite requests; this row reports its exact score field for 1694 requests. Unanswered and unsupported requests count as wrong under the suite methodology.
ToolRet · Decision Index 0.3 (Show measurement, test conditions, and source)
- Source value
- 0.6047
- Score
- 0.6047
- Metric
- nDCG@10
- Unit
- ratio
- Benchmark version
- Decision Index release-v3
- Category
- decision-index-tools
- Direction
- Higher is better
- Sample
- 685
- Source type
- independent-evaluator
- Evaluator
- Jev Decision Index 0.3 public-results data
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Independent evaluation. Jev Decision Index 0.3 release-v3 evaluation of exact model https://huggingface.co/StrandsAgents/strands-decider-2B-hobson-v21. The evaluator sampled one in six public-suite requests; this row reports its exact score field for 685 requests. Unanswered and unsupported requests count as wrong under the suite methodology.
API-Bank · Decision Index 0.3 (Show measurement, test conditions, and source)
- Source value
- 0.3957
- Score
- 0.3957
- Metric
- accuracy
- Unit
- ratio
- Benchmark version
- Decision Index release-v3
- Category
- decision-index-tools
- Direction
- Higher is better
- Sample
- 508
- Source type
- independent-evaluator
- Evaluator
- Jev Decision Index 0.3 public-results data
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Independent evaluation. Jev Decision Index 0.3 release-v3 evaluation of exact model https://huggingface.co/StrandsAgents/strands-decider-2B-hobson-v21. The evaluator sampled one in six public-suite requests; this row reports its exact score field for 508 requests. Unanswered and unsupported requests count as wrong under the suite methodology.
BANKING77 · Decision Index 0.3 (Show measurement, test conditions, and source)
- Source value
- 0.8036
- Score
- 0.8036
- Metric
- macro-F1
- Unit
- ratio
- Benchmark version
- Decision Index release-v3
- Category
- decision-index-retrieval
- Direction
- Higher is better
- Sample
- 3080
- Source type
- independent-evaluator
- Evaluator
- Jev Decision Index 0.3 public-results data
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Independent evaluation. Jev Decision Index 0.3 release-v3 evaluation of exact model https://huggingface.co/StrandsAgents/strands-decider-2B-hobson-v21. The evaluator sampled one in six public-suite requests; this row reports its exact score field for 3080 requests. Unanswered and unsupported requests count as wrong under the suite methodology.
CLINC150 · Decision Index 0.3 (Show measurement, test conditions, and source)
- Source value
- 0.6469
- Score
- 0.6469
- Metric
- macro-F1
- Unit
- ratio
- Benchmark version
- Decision Index release-v3
- Category
- decision-index-retrieval
- Direction
- Higher is better
- Sample
- 5500
- Source type
- independent-evaluator
- Evaluator
- Jev Decision Index 0.3 public-results data
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Independent evaluation. Jev Decision Index 0.3 release-v3 evaluation of exact model https://huggingface.co/StrandsAgents/strands-decider-2B-hobson-v21. The evaluator sampled one in six public-suite requests; this row reports its exact score field for 5500 requests. Unanswered and unsupported requests count as wrong under the suite methodology.
RouterBench · Decision Index 0.3 (Show measurement, test conditions, and source)
- Source value
- 0.7886
- Score
- 0.7886
- Metric
- selected quality (quality objective)
- Unit
- ratio
- Benchmark version
- Decision Index release-v3
- Category
- decision-index-tools
- Direction
- Higher is better
- Sample
- 10000
- Source type
- independent-evaluator
- Evaluator
- Jev Decision Index 0.3 public-results data
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Independent evaluation. Jev Decision Index 0.3 release-v3 evaluation of exact model https://huggingface.co/StrandsAgents/strands-decider-2B-hobson-v21. The evaluator sampled one in six public-suite requests; this row reports its exact score field for 10000 requests. Unanswered and unsupported requests count as wrong under the suite methodology.
- Context
- Not in the index: the prompt reveals that GPT-4 is the best route, so the top models all pick GPT-4 and score within a fraction of a point of each other. Scores stay visible. This score is visible in the raw results but is not included in the Decision Index 0.3 headline score.
Home appliances · Decision Index 0.3 (Show measurement, test conditions, and source)
- Source value
- 0
- Score
- 0
- Metric
- case exact accuracy
- Unit
- ratio
- Benchmark version
- Decision Index release-v3
- Category
- decision-index-tools
- Direction
- Higher is better
- Sample
- 88
- Source type
- independent-evaluator
- Evaluator
- Jev Decision Index 0.3 public-results data
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Independent evaluation. Jev Decision Index 0.3 release-v3 evaluation of exact model https://huggingface.co/StrandsAgents/strands-decider-2B-hobson-v21. The evaluator sampled one in six public-suite requests; this row reports its exact score field for 88 requests. Unanswered and unsupported requests count as wrong under the suite methodology.
SGD · Decision Index 0.3 (Show measurement, test conditions, and source)
- Source value
- 0.5961
- Score
- 0.5961
- Metric
- macro-F1
- Unit
- ratio
- Benchmark version
- Decision Index release-v3
- Category
- decision-index-retrieval
- Direction
- Higher is better
- Sample
- 2500
- Source type
- independent-evaluator
- Evaluator
- Jev Decision Index 0.3 public-results data
- Status
- disputed
- Retrieved at
- 2026-10-07
- Methodology
- Independent evaluation. Jev Decision Index 0.3 release-v3 evaluation of exact model https://huggingface.co/StrandsAgents/strands-decider-2B-hobson-v21. The evaluator sampled one in six public-suite requests; this row reports its exact score field for 2500 requests. Unanswered and unsupported requests count as wrong under the suite methodology.
- Context
- Not in the index: build bug, pending re-run. The builder gave every SGD row the whole dialogue, including the turns after the one being classified. Scores stay visible; SGD returns to the index once re-run with the fixed builder. This score is visible in the raw results but is not included in the Decision Index 0.3 headline score.
ContractNLI · Decision Index 0.3 (Show measurement, test conditions, and source)
- Source value
- 0.7302
- Score
- 0.7302
- Metric
- macro-F1
- Unit
- ratio
- Benchmark version
- Decision Index release-v3
- Category
- decision-index-language
- Direction
- Higher is better
- Sample
- 123
- Source type
- independent-evaluator
- Evaluator
- Jev Decision Index 0.3 public-results data
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Independent evaluation. Jev Decision Index 0.3 release-v3 evaluation of exact model https://huggingface.co/StrandsAgents/strands-decider-2B-hobson-v21. The evaluator sampled one in six public-suite requests; this row reports its exact score field for 123 requests. Unanswered and unsupported requests count as wrong under the suite methodology.
ANLI · Decision Index 0.3 (Show measurement, test conditions, and source)
- Source value
- 0.399
- Score
- 0.399
- Metric
- macro-F1
- Unit
- ratio
- Benchmark version
- Decision Index release-v3
- Category
- decision-index-language
- Direction
- Higher is better
- Sample
- 3200
- Source type
- independent-evaluator
- Evaluator
- Jev Decision Index 0.3 public-results data
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Independent evaluation. Jev Decision Index 0.3 release-v3 evaluation of exact model https://huggingface.co/StrandsAgents/strands-decider-2B-hobson-v21. The evaluator sampled one in six public-suite requests; this row reports its exact score field for 3200 requests. Unanswered and unsupported requests count as wrong under the suite methodology.
BPoMP · Decision Index 0.3 (Show measurement, test conditions, and source)
- Source value
- 0.56
- Score
- 0.56
- Metric
- accuracy
- Unit
- ratio
- Benchmark version
- Decision Index release-v3
- Category
- decision-index-arts
- Direction
- Higher is better
- Sample
- 5000
- Source type
- independent-evaluator
- Evaluator
- Jev Decision Index 0.3 public-results data
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Independent evaluation. Jev Decision Index 0.3 release-v3 evaluation of exact model https://huggingface.co/StrandsAgents/strands-decider-2B-hobson-v21. The evaluator sampled one in six public-suite requests; this row reports its exact score field for 5000 requests. Unanswered and unsupported requests count as wrong under the suite methodology.
Humicroedit · Decision Index 0.3 (Show measurement, test conditions, and source)
- Source value
- 0.4973
- Score
- 0.4973
- Metric
- accuracy
- Unit
- ratio
- Benchmark version
- Decision Index release-v3
- Category
- decision-index-arts
- Direction
- Higher is better
- Sample
- 2628
- Source type
- independent-evaluator
- Evaluator
- Jev Decision Index 0.3 public-results data
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Independent evaluation. Jev Decision Index 0.3 release-v3 evaluation of exact model https://huggingface.co/StrandsAgents/strands-decider-2B-hobson-v21. The evaluator sampled one in six public-suite requests; this row reports its exact score field for 2628 requests. Unanswered and unsupported requests count as wrong under the suite methodology.
Page 1 of 7
Sources and data date
| Type | Evidence and data date |
|---|---|
| Research status | Research date October 7, 2026. 4 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Additional source | Strands Decider launch announcement (retrieved October 7, 2026) · Strands Decider launch announcement · Editorial description reviewed October 7, 2026 |
| Additional source | Strands Decider v21 model card (retrieved October 7, 2026) · Strands Decider v21 model card · Editorial description reviewed October 7, 2026 |
| Additional source | Strands Decider v21 release commit (retrieved October 7, 2026) |
| Additional source | Jev Decision Index 0.3 public-results data (retrieved October 7, 2026) |
| Additional source | Jev Decision Index 0.3 score decomposition data (retrieved October 7, 2026) |
| Additional source | Strands Decider v21 model card (retrieved October 7, 2026) |