# Bison 1

> Bison 1 is an open-weight Base Labs decision model post-trained from Gemma 4 31B. It returns calibrated probabilities for choice, yes/no and score questions over text and images, with a 128K-token context and the top score on the independent decision index v2.1.

**ID**: `baselabs/bison-1`  
**Creator**: baselabs  
**Category**: text  
**Context**: 131K tokens  
**Pricing**: $0.10 / $0 per 1M · billed by your provider (BYOK)  
**Released**: 2026-11-20  
**Links**: [Weights on Hugging Face](https://huggingface.co/baselabs/bison-1) · [Announcement](https://baseten.co/blog/meet-bison-1-and-groundhog-1/) · [AnyRouter release](https://anyrouter.dev/blog/releases/2026-11-21-bison-1)  
**Web page**: https://anyrouter.dev/model/baselabs/bison-1

**Input modalities**: text, image  
**Output modalities**: text  

**Capabilities**: systemone

**Supported parameters**: state, model, questions, service_tier, max_len

## Usage

**Endpoint**: `POST https://anyrouter.dev/api/v1/decisions`

```bash
curl https://anyrouter.dev/api/v1/decisions \
  -H "Authorization: Bearer $ANYROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "state": "Help! My payouts have been failing for 3 days.",
  "model": "baselabs/bison-1",
  "questions": {
    "is_urgent": {
      "type": "noul",
      "instructions": "Does this convey urgency?"
    }
  }
}'
```

Decisions is not chat: send `state` and typed `questions` (noul / choice / score). Requires a BYOK key for TypeSafe, LLM Gateway, Vercel AI Gateway, AIHubMix, AutoJev, or Fastino.

## Providers

| Provider | Input | Output |
| --- | --- | --- |
| baseten-systemone-byok | $0.10/M (list) | $0 (list) |

BYOK routes are billed by your own provider key — AnyRouter charges $0. Rates marked (list) are the provider's published list price.

## Model card

- Architecture: dense [2]
- Total parameters: 31B [2]
- Native context: 131072 tokens [2,1]
- License: MIT, open weights (https://huggingface.co/baselabs/bison-1) [3,2]

### Strengths [1]

- Ranked first of 19 decision models on the independent decision index v2.1 with a score of 76, ahead of d1.2-ultra at 75 and mammoth-1 at 71.
- Lowest Brier score on the index at 0.14, so its stated confidence tracks how often it is right.
- 128K context, eight times the previous generation, with text and image input.

### Known limits [2]

- Returns structured decisions and option probabilities, not free-form chat text.
- Designed for classification, routing and scoring; it does not generate long-form answers.

### Use cases [2]

- Routing requests and classifying cases over long documents or policy text.
- Yes/no checks and scoring against a rubric with calibrated confidence thresholds.
- Decisions over text and images using choice, noul and score questions.

### API notes [4,2]

- Native System One: POST https://inference.baseten.co/v1/systemone with {model, state, questions}; anyrouter exposes it at /v1/systemone.
- Auth is a Baseten API key sent as a Bearer token; add it under Dashboard → BYOK.
- Questions are choice, noul (yes/no) or score; Baseten bills input tokens per 1M and output is one token per question.

### Links

- announcement: https://baseten.co/blog/meet-bison-1-and-groundhog-1/
- model card: https://www.baseten.co/library/bison-1/
- docs: https://docs.baseten.co/reference/inference-api/system-one.md

### Sources

1. [Meet bison-1 and groundhog-1, frontier American decision models](https://baseten.co/blog/meet-bison-1-and-groundhog-1/) (accessed 2026-11-21)
2. [bison-1 model page - Baseten](https://www.baseten.co/library/bison-1/) (accessed 2026-11-21)
3. [baselabs/bison-1 released weights and MIT license](https://huggingface.co/baselabs/bison-1) (accessed 2026-11-21)
4. [Baseten System One API reference](https://docs.baseten.co/reference/inference-api/system-one.md) (accessed 2026-11-21)

## Source

- Upstream docs: https://baseten.co/blog/meet-bison-1-and-groundhog-1/
