Decision modelOpen source

Intern-Decision-4B

InternLM

Released: -
Data date: October 3, 2026

Model class: Multimodal structured decision model, not a free-text generator
Access: Local Hugging Face execution with the bundled DecisionEngine

Intern-Decision-4B is the largest of the three Intern-Decision variants. The multimodal model reads shared state, named questions, and optional images. A causal forward pass scores the allowed answer symbols and turns them into typed JSON answers with Choice, Score, or Noul, without calling generate.

The checkpoint is fine-tuned from Qwen3.5-4B, runs locally with the DecisionEngine, and accepts up to 8,192 input tokens; the release is under Apache-2.0.

Specifications and access

Specification	Value and source
Model class	Multimodal structured decision model, not a free-text generatorSource
Output schema	Choice, Score, and Noul as typed JSON answers with probabilitiesSource
Input	State, named questions, and optional imagesSource
Access	Local Hugging Face execution with the bundled DecisionEngineSource
Checkpoint	internlm/Intern-Decision-4BSource
Base model	Qwen/Qwen3.5-4BSource
Context window	8,192-token runtime input limitSource
License	Apache-2.0 with the Qwen notices retainedSource

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.

Intern Decision evaluation: Jevbench-Easy (Show measurement, test conditions, and source)

Source value: 100
Score: 100
Metric: Jevbench-Easy
Unit: %
Category: Intern Decision evaluation
Direction: Higher is better
Source type: vendor-reported
Evaluator: InternLM Intern-Decision-4B model card
Status: active
Retrieved at: 2026-10-04
Methodology: Vendor-reported result. Local Hugging Face inference=yes; Structured candidate scoring=yes

Source InternLM Intern-Decision-4B model card

Intern Decision evaluation: Jevbench-Original (Show measurement, test conditions, and source)

Source value: 98.61
Score: 98.61
Metric: Jevbench-Original
Unit: %
Category: Intern Decision evaluation
Direction: Higher is better
Source type: vendor-reported
Evaluator: InternLM Intern-Decision-4B model card
Status: active
Retrieved at: 2026-10-04
Methodology: Vendor-reported result. Local Hugging Face inference=yes; Structured candidate scoring=yes

Source InternLM Intern-Decision-4B model card

Intern Decision evaluation: Jevbench-Hard (Show measurement, test conditions, and source)

Source value: 73.87
Score: 73.87
Metric: Jevbench-Hard
Unit: %
Category: Intern Decision evaluation
Direction: Higher is better
Source type: vendor-reported
Evaluator: InternLM Intern-Decision-4B model card
Status: active
Retrieved at: 2026-10-04
Methodology: Vendor-reported result. Local Hugging Face inference=yes; Structured candidate scoring=yes

Source InternLM Intern-Decision-4B model card

Intern Decision evaluation: Typed Decision (Show measurement, test conditions, and source)

Source value: 80.55
Score: 80.55
Metric: Typed Decision
Unit: %
Category: Intern Decision evaluation
Direction: Higher is better
Source type: vendor-reported
Evaluator: InternLM Intern-Decision-4B model card
Status: active
Retrieved at: 2026-10-04
Methodology: Vendor-reported result. Local Hugging Face inference=yes; Structured candidate scoring=yes

Source InternLM Intern-Decision-4B model card

Intern Decision evaluation: ToolACE (Show measurement, test conditions, and source)

Source value: 96.45
Score: 96.45
Metric: ToolACE
Unit: %
Category: Intern Decision evaluation
Direction: Higher is better
Source type: vendor-reported
Evaluator: InternLM Intern-Decision-4B model card
Status: active
Retrieved at: 2026-10-04
Methodology: Vendor-reported result. Local Hugging Face inference=yes; Structured candidate scoring=yes

Source InternLM Intern-Decision-4B model card

Intern Decision evaluation: AG News (Show measurement, test conditions, and source)

Source value: 90.82
Score: 90.82
Metric: AG News
Unit: %
Category: Intern Decision evaluation
Direction: Higher is better
Source type: vendor-reported
Evaluator: InternLM Intern-Decision-4B model card
Status: active
Retrieved at: 2026-10-04
Methodology: Vendor-reported result. Local Hugging Face inference=yes; Structured candidate scoring=yes

Source InternLM Intern-Decision-4B model card

Intern Decision evaluation: WildJailBreak (Show measurement, test conditions, and source)

Source value: 89.86
Score: 89.86
Metric: WildJailBreak
Unit: %
Category: Intern Decision evaluation
Direction: Higher is better
Source type: vendor-reported
Evaluator: InternLM Intern-Decision-4B model card
Status: active
Retrieved at: 2026-10-04
Methodology: Vendor-reported result. Local Hugging Face inference=yes; Structured candidate scoring=yes

Source InternLM Intern-Decision-4B model card

Intern Decision evaluation: Average (Show measurement, test conditions, and source)

Source value: 90.02
Score: 90.02
Metric: Average
Unit: %
Category: Intern Decision evaluation
Direction: Higher is better
Source type: vendor-reported
Evaluator: InternLM Intern-Decision-4B model card
Status: active
Retrieved at: 2026-10-04
Methodology: Vendor-reported result. Local Hugging Face inference=yes; Structured candidate scoring=yes

Source InternLM Intern-Decision-4B model card

Intern Decision evaluation: Brier (Show measurement, test conditions, and source)

Source value: 0.347
Score: 0.347
Metric: Brier
Unit: ratio
Category: Intern Decision evaluation
Direction: Lower is better
Source type: vendor-reported
Evaluator: InternLM Intern-Decision-4B model card
Status: active
Retrieved at: 2026-10-04
Methodology: Vendor-reported result. Local Hugging Face inference=yes; Structured candidate scoring=yes

Source InternLM Intern-Decision-4B model card

Intern Decision evaluation: ECE (Show measurement, test conditions, and source)

Source value: 0.065
Score: 0.065
Metric: ECE
Unit: ratio
Category: Intern Decision evaluation
Direction: Lower is better
Source type: vendor-reported
Evaluator: InternLM Intern-Decision-4B model card
Status: active
Retrieved at: 2026-10-04
Methodology: Vendor-reported result. Local Hugging Face inference=yes; Structured candidate scoring=yes

Source InternLM Intern-Decision-4B model card

Intern Decision inference latency: Mean (seconds) (Show measurement, test conditions, and source)

Source value: 0.04416
Score: 0.04416
Metric: Mean
Unit: seconds
Category: Intern Decision inference latency
Direction: Lower is better
Source type: vendor-reported
Evaluator: InternLM Intern-Decision-4B model card
Status: active
Retrieved at: 2026-10-04
Methodology: Vendor-reported result. Hardware=Single RTX 4090; Inference path=local HF; Time definition=Per-query end-to-end latency

Source InternLM Intern-Decision-4B model card

Intern Decision inference latency: Median / P50 (seconds) (Show measurement, test conditions, and source)

Source value: 0.04403
Score: 0.04403
Metric: Median / P50
Unit: seconds
Category: Intern Decision inference latency
Direction: Lower is better
Source type: vendor-reported
Evaluator: InternLM Intern-Decision-4B model card
Status: active
Retrieved at: 2026-10-04
Methodology: Vendor-reported result. Hardware=Single RTX 4090; Inference path=local HF; Time definition=Per-query end-to-end latency

Source InternLM Intern-Decision-4B model card

Page 1 of 3

Sources and data date

Every statement links to its underlying documentation or leaderboard.

Type	Evidence and data date
Research status	Research date October 4, 2026. 1 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication.
Additional source	Intern-Decision collection (retrieved October 3, 2026) · Intern-Decision collection · Editorial description reviewed October 3, 2026
Additional source	Intern-Decision repository (retrieved October 3, 2026) · Intern-Decision repository · Editorial description reviewed October 3, 2026
Additional source	Intern-Decision-4B model card (retrieved October 3, 2026; October 4, 2026) · Intern-Decision-4B model card · Editorial description reviewed October 3, 2026