Skip to main content

BasetenDecision modelOpen weights and API

Groundhog 1

Summary

Groundhog 1 is Baseten's mixture-of-experts decision model for structured Choice and Score outputs. It builds on Gemma 4 26B A4B IT and processes text or images. Four billion of its 26 billion parameters are active.

All benchmark results

Every published result with its source, test setup and comparison models.

Show all 6 published benchmark results

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.

ClaimBench: Direct support (Show measurement, test conditions, and source)
Source value
100
Metric
accuracy
Unit
%
Benchmark version
1
Category
finance
Task
direct_support
Task label
Direct support
Direction
Higher is better
Source type
independent-evaluator
Evaluator
Vals AI
Status
active
Retrieved at
2026-11-21
Methodology
Source model ID: decision/groundhog-1. Source model name: Groundhog 1. Provider reported by source: Baseten. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-11-20. Maximum output tokens: 0. Latency in seconds: 0.729. Standard error: 0. Cost per task in USD: 0.000175. Source ID: vals-ai. Evidence capture ID: vals-claimbench-2026-11-21-5dc3eba20db7. Source payload hash: 5dc3eba20db74e5214c71d077ff3785a5201c2ca4303917a1b122f03665d8562. Parser version: vals-astro-v4. Imported at: 2026-11-21. Source cohort: vals-claimbench:1:direct_support
Source Vals AI
ClaimBench: Missing evidence (Show measurement, test conditions, and source)
Source value
100
Metric
accuracy
Unit
%
Benchmark version
1
Category
finance
Task
missing_evidence
Task label
Missing evidence
Direction
Higher is better
Source type
independent-evaluator
Evaluator
Vals AI
Status
active
Retrieved at
2026-11-21
Methodology
Source model ID: decision/groundhog-1. Source model name: Groundhog 1. Provider reported by source: Baseten. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-11-20. Maximum output tokens: 0. Latency in seconds: 0.814. Standard error: 0. Cost per task in USD: 0.000185. Source ID: vals-ai. Evidence capture ID: vals-claimbench-2026-11-21-5dc3eba20db7. Source payload hash: 5dc3eba20db74e5214c71d077ff3785a5201c2ca4303917a1b122f03665d8562. Parser version: vals-astro-v4. Imported at: 2026-11-21. Source cohort: vals-claimbench:1:missing_evidence
Source Vals AI
ClaimBench: Non-supporting citation (Show measurement, test conditions, and source)
Source value
100
Metric
accuracy
Unit
%
Benchmark version
1
Category
finance
Task
non_supporting_citation
Task label
Non-supporting citation
Direction
Higher is better
Source type
independent-evaluator
Evaluator
Vals AI
Status
active
Retrieved at
2026-11-21
Methodology
Source model ID: decision/groundhog-1. Source model name: Groundhog 1. Provider reported by source: Baseten. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-11-20. Maximum output tokens: 0. Latency in seconds: 0.729. Standard error: 0. Cost per task in USD: 0.000184. Source ID: vals-ai. Evidence capture ID: vals-claimbench-2026-11-21-5dc3eba20db7. Source payload hash: 5dc3eba20db74e5214c71d077ff3785a5201c2ca4303917a1b122f03665d8562. Parser version: vals-astro-v4. Imported at: 2026-11-21. Source cohort: vals-claimbench:1:non_supporting_citation
Source Vals AI
ClaimBench: Numeric discrepancy (Show measurement, test conditions, and source)
Source value
100
Metric
accuracy
Unit
%
Benchmark version
1
Category
finance
Task
numeric_discrepancy
Task label
Numeric discrepancy
Direction
Higher is better
Source type
independent-evaluator
Evaluator
Vals AI
Status
active
Retrieved at
2026-11-21
Methodology
Source model ID: decision/groundhog-1. Source model name: Groundhog 1. Provider reported by source: Baseten. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-11-20. Maximum output tokens: 0. Latency in seconds: 0.746. Standard error: 0. Cost per task in USD: 0.000188. Source ID: vals-ai. Evidence capture ID: vals-claimbench-2026-11-21-5dc3eba20db7. Source payload hash: 5dc3eba20db74e5214c71d077ff3785a5201c2ca4303917a1b122f03665d8562. Parser version: vals-astro-v4. Imported at: 2026-11-21. Source cohort: vals-claimbench:1:numeric_discrepancy
Source Vals AI
ClaimBench: Overall (Show measurement, test conditions, and source)
Source value
99.031
Metric
accuracy
Unit
%
Benchmark version
1
Category
finance
Task
overall
Task label
Overall
Direction
Higher is better
Source type
independent-evaluator
Evaluator
Vals AI
Status
active
Retrieved at
2026-11-21
Methodology
Source model ID: decision/groundhog-1. Source model name: Groundhog 1. Provider reported by source: Baseten. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-11-20. Maximum output tokens: 0. Latency in seconds: 0.759. Standard error: 0.221. Cost per task in USD: 0.000181. Source ID: vals-ai. Evidence capture ID: vals-claimbench-2026-11-21-5dc3eba20db7. Source payload hash: 5dc3eba20db74e5214c71d077ff3785a5201c2ca4303917a1b122f03665d8562. Parser version: vals-astro-v4. Imported at: 2026-11-21. Source cohort: vals-claimbench:1:overall
Source Vals AI
ClaimBench: Subtle contradiction (Show measurement, test conditions, and source)
Source value
95.153
Metric
accuracy
Unit
%
Benchmark version
1
Category
finance
Task
subtle_contradiction
Task label
Subtle contradiction
Direction
Higher is better
Source type
independent-evaluator
Evaluator
Vals AI
Status
active
Retrieved at
2026-11-21
Methodology
Source model ID: decision/groundhog-1. Source model name: Groundhog 1. Provider reported by source: Baseten. Evaluator: Vals AI. Dataset type: public. Leaderboard date: 2026-11-20. Maximum output tokens: 0. Latency in seconds: 0.776. Standard error: 1.086. Cost per task in USD: 0.000175. Source ID: vals-ai. Evidence capture ID: vals-claimbench-2026-11-21-5dc3eba20db7. Source payload hash: 5dc3eba20db74e5214c71d077ff3785a5201c2ca4303917a1b122f03665d8562. Parser version: vals-astro-v4. Imported at: 2026-11-21. Source cohort: vals-claimbench:1:subtle_contradiction
Source Vals AI

All specifications

Architecture

Model class
A structured decision model for Choice and Score outputs, not a free-form text generator.
Source: Model class
Base model
Gemma 4 26B A4B IT
Source: Base model
Architecture
Mixture of experts with 26 billion total parameters and 4 billion active parameters.
Source: Architecture

Input and output

Inputs and outputs
Text and images in. Structured decision values out.
Source: Inputs and outputs
Context window
131,072 tokens
Source: Context window

Watermarking and provenance

AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.

AI watermarking
Unknown
no source
Watermarking technique
Not known
no source

Licence and access

License
Baseten's model modifications are under MIT. Google's Apache 2.0 terms apply to the underlying Gemma weights.
Source: License
Shutdown
None announced
no source

API

Baseten API model ID
decision-models/baseten-groundhog-latest
Source: Baseten API model ID

Further details

Checkpoint
baselabs/groundhog-1 (BF16)
Source: Checkpoint

Prices

Baseten API price
$0.035 per 1M tokensSource

Sources and data date

Benchmark sources at a glance

Which independent evaluators tested this model. This is coverage, not a quality ranking.

EvaluatorTest seriesComposite indicesRecorded active measurements
Vals AI106

Without vendor claims and historical or disputed results; those stay labelled in the single values.