Skip to main content

InternLMDecision modelOpen source

Intern-Decision-2B

PopularityRank 20 of 39 decision models

31out of 100new calculation

based on two data groups, mostly search, range 21 to 41

Summary

Intern-Decision-2B is the middle multimodal variant in the Intern-Decision line. The checkpoint processes shared state, a schema of named questions, and optional images. It computes a typed JSON answer for each field in one forward pass, using Choice, Score, or Noul, and does not return free-form text.

Its two billion parameters come from fine-tuning Qwen3.5-2B. The local Apache-2.0 model uses the DecisionEngine and accepts up to 8,192 input tokens.

GraduallyBench Decision

Intern-Decision-2B decides 39 of 60 test cases correctly. The median of the 33 tested decision models is 60.0 %.

Accuracy

65.0 %

39 of 60 test cases

Robustness

73.3 %

22 of 30 reworded variants

Unnecessary reviews

5.8 %

3 of 52 cases without a required review

Response time

290 ms

median, 95 % under 575 ms, local, GPU

Accuracy compared

Share of correct decisions in percent, higher is better

20 of 33 models

Jev93.3 %
Kev 9B93.3 %
Clef91.7 %
Clef-Flash85.0 %
Decider 4B85.0 %
d1-2B85.0 %
Lev81.7 %
d1-Omni-2B81.7 %
Kev 4B78.3 %
Jeff 170.0 %
Decider 2B66.7 %
Intern-Decision-2B65.0 %
Kev 0.8B50.0 %

As of 11/10/2026. Source: GraduallyBench Decision

Accuracy by task familyThe tick marks the median of all 33 tested models.
Triage50.0 %6 / 12, median 58.3 %
Routing91.7 %11 / 12, median 83.3 %
Constraints25.0 %3 / 12, median 33.3 %
Tables66.7 %8 / 12, median 50.0 %
Evidence91.7 %11 / 12, median 75.0 %

Intern-Decision-2B

Benchmark run
Show test details

Decisions

Local · German, English

Execution Apple Metal. Precision FP32.

130 valid outputs from 130 attempted and 130 planned test slots.

Measured duration p50 / p95 290 / 575 ms (130 cases). Local GPU. Quantile calculation not documented.

Automatic decision accuracy
65 %(39 / 60 planned cases)German, English
Automatic decision accuracy, Triage
50 %(6 / 12 planned cases)German, English
Automatic decision accuracy, Routing
91.67 %(11 / 12 planned cases)German, English
Automatic decision accuracy, Selection with constraints
25 %(3 / 12 planned cases)German, English
Automatic decision accuracy, Numbers and tables
66.67 %(8 / 12 planned cases)German, English
Automatic decision accuracy, Sufficient evidence
91.67 %(11 / 12 planned cases)German, English
Correct robustness variants
73.33 %(22 / 30 planned cases)German, English
Correct development cases
77.5 %(31 / 40 planned cases)German, English
Unnecessary reviews
5.77 %(3 / 52 assessable cases without a required review)German, English
Missed required reviews
25 %(2 / 8 assessable cases requiring review)German, English

Response times are comparable only within the same environment and form no shared speed ranking. Where used, the test references were not independently reviewed by humans. Test coverage and methods differ by model. Each metric stands on its own; there is no combined overall score across tests. Data compiled on 11/10/2026. How we test · Download all Decision and Audio results as JSON

All benchmark results

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

Show all 58 published benchmark results

Page 1 of 5

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
Metric
Jevbench-Easy
Unit
%
Benchmark version
arXiv:2610.22131v1, Table 3, Jevbench-Easy
Category
Intern Decision evaluation
Direction
Higher is better
Sample
48
Source type
vendor-reported
Evaluator
InternLM Intern-Decision-2B model card
Status
active
Retrieved at
2026-10-04
Methodology
Vendor-reported result. Table 3: local Hugging Face inference with structured candidate scoring. Exact model label: Intern-Decision-2B. Evaluation suite: Jevbench-Easy. Table 2 reports N=48 decisions for this suite.
Source InternLM Intern-Decision-2B model card
Intern Decision evaluation: Jevbench-Original (Show measurement, test conditions, and source)
Source value
84.72
Metric
Jevbench-Original
Unit
%
Benchmark version
arXiv:2610.22131v1, Table 3, Jevbench-Original
Category
Intern Decision evaluation
Direction
Higher is better
Sample
72
Source type
vendor-reported
Evaluator
InternLM Intern-Decision-2B model card
Status
active
Retrieved at
2026-10-04
Methodology
Vendor-reported result. Table 3: local Hugging Face inference with structured candidate scoring. Exact model label: Intern-Decision-2B. Evaluation suite: Jevbench-Original. Table 2 reports N=72 decisions for this suite.
Source InternLM Intern-Decision-2B model card
Intern Decision evaluation: Jevbench-Hard (Show measurement, test conditions, and source)
Source value
63.96
Metric
Jevbench-Hard
Unit
%
Benchmark version
arXiv:2610.22131v1, Table 3, Jevbench-Hard
Category
Intern Decision evaluation
Direction
Higher is better
Sample
111
Source type
vendor-reported
Evaluator
InternLM Intern-Decision-2B model card
Status
active
Retrieved at
2026-10-04
Methodology
Vendor-reported result. Table 3: local Hugging Face inference with structured candidate scoring. Exact model label: Intern-Decision-2B. Evaluation suite: Jevbench-Hard. Table 2 reports N=111 decisions for this suite.
Source InternLM Intern-Decision-2B model card
Intern Decision evaluation: Typed Decision (Show measurement, test conditions, and source)
Source value
79.35
Metric
Typed Decision
Unit
%
Benchmark version
arXiv:2610.22131v1, Table 3, Typed Decision
Category
Intern Decision evaluation
Direction
Higher is better
Sample
2000
Source type
vendor-reported
Evaluator
InternLM Intern-Decision-2B model card
Status
active
Retrieved at
2026-10-04
Methodology
Vendor-reported result. Table 3: local Hugging Face inference with structured candidate scoring. Exact model label: Intern-Decision-2B. Evaluation suite: Typed Decision. Table 2 reports N=2,000 decisions for this suite.
Source InternLM Intern-Decision-2B model card
Intern Decision evaluation: ToolACE (Show measurement, test conditions, and source)
Source value
96.45
Metric
ToolACE
Unit
%
Benchmark version
arXiv:2610.22131v1, Table 3, ToolACE
Category
Intern Decision evaluation
Direction
Higher is better
Sample
310
Source type
vendor-reported
Evaluator
InternLM Intern-Decision-2B model card
Status
active
Retrieved at
2026-10-04
Methodology
Vendor-reported result. Table 3: local Hugging Face inference with structured candidate scoring. Exact model label: Intern-Decision-2B. Evaluation suite: ToolACE. Table 2 reports N=310 decisions for this suite.
Source InternLM Intern-Decision-2B model card
Intern Decision evaluation: AG News (Show measurement, test conditions, and source)
Source value
89.96
Metric
AG News
Unit
%
Benchmark version
arXiv:2610.22131v1, Table 3, AG News
Category
Intern Decision evaluation
Direction
Higher is better
Sample
7600
Source type
vendor-reported
Evaluator
InternLM Intern-Decision-2B model card
Status
active
Retrieved at
2026-10-04
Methodology
Vendor-reported result. Table 3: local Hugging Face inference with structured candidate scoring. Exact model label: Intern-Decision-2B. Evaluation suite: AG News. Table 2 reports N=7,600 decisions for this suite.
Source InternLM Intern-Decision-2B model card
Intern Decision evaluation: WildJailBreak (Show measurement, test conditions, and source)
Source value
78.33
Metric
WildJailBreak
Unit
%
Benchmark version
arXiv:2610.22131v1, Table 3, WildJailBreak
Category
Intern Decision evaluation
Direction
Higher is better
Sample
2210
Source type
vendor-reported
Evaluator
InternLM Intern-Decision-2B model card
Status
active
Retrieved at
2026-10-04
Methodology
Vendor-reported result. Table 3: local Hugging Face inference with structured candidate scoring. Exact model label: Intern-Decision-2B. Evaluation suite: WildJailBreak. Table 2 reports N=2,210 decisions for this suite.
Source InternLM Intern-Decision-2B model card
Intern Decision evaluation: Average (Show measurement, test conditions, and source)
Source value
84.68
Metric
Average
Unit
%
Category
Intern Decision evaluation
Direction
Higher is better
Source type
vendor-reported
Evaluator
InternLM Intern-Decision-2B 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-2B model card
Zero-shot IE multilingual: popular-language tier accuracy (Show measurement, test conditions, and source)
Source value
100
Metric
accuracy, source table rounded to whole percent
Unit
%
Benchmark version
Zero-shot IE multilingual snapshot at repository commit c45a00c25d2731ac7047def97b91421aaff2f611
Category
multilingual-popular
Direction
Higher is better
Sample
18
Source type
independent-evaluator
Evaluator
Zero-shot IE benchmark README, pinned commit c45a00c25d2731ac7047def97b91421aaff2f611
Status
active
Retrieved at
2026-10-17
Methodology
Independent evaluation. The pinned author README's multilingual table reports the Popular tier for exact result row “Intern-Decision 2B”: 100%. The task comprises six sentiment texts per each of nine languages; this tier includes Spanish, French and Simplified Chinese, 18 texts total. The result cell is taken directly from the pinned README table (body SHA-256 7905d1cbfb4d75d65d02a5e6c537e1e84fe8b5c388f33075a6c0db42fa0cac98), not recomputed from the separate language scores. The public runner describes blind back-translation; it uses label options with English descriptions. No immutable model/API revision or confidence interval is published.
Context
The pinned author README publishes rounded whole-percent tier cells; these exact displayed cells are used, not recomputed from the language-level data. This is a custom, small evaluation rather than a standardized leaderboard. The author reports blind back-translation and says the full text table was in a PR for native-speaker review; this review itself was not independently confirmed. System/model revisions and full decoding settings are not pinned.
Source Zero-shot IE benchmark README, pinned commit c45a00c25d2731ac7047def97b91421aaff2f611
Zero-shot IE multilingual: medium-resource-language tier accuracy (Show measurement, test conditions, and source)
Source value
100
Metric
accuracy, source table rounded to whole percent
Unit
%
Benchmark version
Zero-shot IE multilingual snapshot at repository commit c45a00c25d2731ac7047def97b91421aaff2f611
Category
multilingual-medium
Direction
Higher is better
Sample
18
Source type
independent-evaluator
Evaluator
Zero-shot IE benchmark README, pinned commit c45a00c25d2731ac7047def97b91421aaff2f611
Status
active
Retrieved at
2026-10-17
Methodology
Independent evaluation. The pinned author README's multilingual table reports the Medium tier for exact result row “Intern-Decision 2B”: 100%. The task comprises six sentiment texts per each of nine languages; this tier includes Vietnamese, Turkish and Ukrainian, 18 texts total. The result cell is taken directly from the pinned README table (body SHA-256 7905d1cbfb4d75d65d02a5e6c537e1e84fe8b5c388f33075a6c0db42fa0cac98), not recomputed from the separate language scores. The public runner describes blind back-translation; it uses label options with English descriptions. No immutable model/API revision or confidence interval is published.
Context
The pinned author README publishes rounded whole-percent tier cells; these exact displayed cells are used, not recomputed from the language-level data. This is a custom, small evaluation rather than a standardized leaderboard. The author reports blind back-translation and says the full text table was in a PR for native-speaker review; this review itself was not independently confirmed. System/model revisions and full decoding settings are not pinned.
Source Zero-shot IE benchmark README, pinned commit c45a00c25d2731ac7047def97b91421aaff2f611
Zero-shot IE multilingual: rare-language tier accuracy (Show measurement, test conditions, and source)
Source value
72
Metric
accuracy, source table rounded to whole percent
Unit
%
Benchmark version
Zero-shot IE multilingual snapshot at repository commit c45a00c25d2731ac7047def97b91421aaff2f611
Category
multilingual-rare
Direction
Higher is better
Sample
18
Source type
independent-evaluator
Evaluator
Zero-shot IE benchmark README, pinned commit c45a00c25d2731ac7047def97b91421aaff2f611
Status
active
Retrieved at
2026-10-17
Methodology
Independent evaluation. The pinned author README's multilingual table reports the Rare tier for exact result row “Intern-Decision 2B”: 72%. The task comprises six sentiment texts per each of nine languages; this tier includes Sinhala, Icelandic and Welsh, 18 texts total. The result cell is taken directly from the pinned README table (body SHA-256 7905d1cbfb4d75d65d02a5e6c537e1e84fe8b5c388f33075a6c0db42fa0cac98), not recomputed from the separate language scores. The public runner describes blind back-translation; it uses label options with English descriptions. No immutable model/API revision or confidence interval is published.
Context
The pinned author README publishes rounded whole-percent tier cells; these exact displayed cells are used, not recomputed from the language-level data. This is a custom, small evaluation rather than a standardized leaderboard. The author reports blind back-translation and says the full text table was in a PR for native-speaker review; this review itself was not independently confirmed. System/model revisions and full decoding settings are not pinned.
Source Zero-shot IE benchmark README, pinned commit c45a00c25d2731ac7047def97b91421aaff2f611
JEVal (arXiv:2610.03935v2), overall choice/noul evaluation (Show measurement, test conditions, and source)
Source value
55.72
Metric
accuracy (invalid outputs count as incorrect)
Unit
%
Benchmark version
arXiv:2610.03935v2, Table 1
Category
JEVal overall choice/noul
Direction
Higher is better
Sample
10938
Source type
community-reported
Evaluator
Duan et al., General Decision Models: Benchmarking and Insights Beyond Jev (arXiv:2610.03935v2)
Status
active
Retrieved at
2026-10-18
Methodology
Community-reported result. JEVal Table 1, exact reported model label Intern-Decision-2B; decision-model run, no reasoning-effort setting reported; local vLLM on NVIDIA H100; context window 32,768 tokens; input budget 24,576 tokens; temperature 2.1005; 10,938 choice and noul questions.
Context
No immutable evaluated-weight or serving-revision digest is given. Table 1 counts invalid outputs as incorrect for accuracy; ECE and raw Brier are computed only on valid responses, whose per-model N is not reported. The paper reports no confidence interval for these point values.
Source Duan et al., General Decision Models: Benchmarking and Insights Beyond Jev (arXiv:2610.03935v2)

Page 1 of 5

Popularity compared

How often the model is searched for, used and mentioned, compared with every model of its category.

Popularity index

0 to 100, higher is more popular

20 of 42 models

As of November 2026 · Source: gradually.ai popularity index

Over time

Weekly values from October 18, 2026 to November 20, 2026

Intern-Decision-2Bpplx-decider-v1.2 33pplx-decider-flash-v1.2 32GLiNER2.5-multi-Decide 30pplx-decider-flash-v1.3 30

Method changes in the timeline, a jump at these points is no market event:

  1. 1November 1, 2026: Value only with search or usage data, 30-day attention
  2. 2November 1, 2026: Method version 4 (common scale, shrinkage)
  3. 3November 9, 2026: New topics measurement V20 (daily quorum, short forms)

What is gaining popularity right now, across models, vendors, harnesses and AI subscriptions, is collected in All AI trends.

All specifications

Architecture

Model class
Multimodal structured decision model, not a free-text generator
Source: Model class
Checkpoint
internlm/Intern-Decision-2B
Source: Checkpoint
Base model
Qwen/Qwen3.5-2B
Source: Base model

Input and output

Input
State, named questions, and optional images
Source: Input
Context window
8,192-token runtime input limit
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

Access
Local Hugging Face execution with the bundled DecisionEngine
Source: Access
License
Apache-2.0 with the Qwen notices retained
Source: License
Shutdown
None announced
no source

Further details

Output schema
Choice, Score, and Noul as typed JSON answers with probabilities
Source: Output schema

Sources and data date

Show all 17 sources

Benchmark sources at a glance

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

No active result from the reviewed external evaluators can currently be attributed to this model. Other source evidence remains available below.

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

TypeEvidence and data date
Research statusResearch date October 17, 2026. 9 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication.
Not published

immutable-model-or-route-revision

Note

The zero-shot IE custom benchmark maps its exact public label to the profile task, but does not publish the loaded weight hash or immutable serving revision.

Not published

sampling-uncertainty

Note

The zero-shot IE small custom pools have no published confidence intervals. This gap is specific to those score cells.

Not published

Immutable Decision Index served checkpoint revision

Note

The current public index gives a model URL and, where available, a served_checkpoint label, but no immutable commit/digest for the exact evaluated weights.

Additional sourceIntern-Decision collection (retrieved October 3, 2026) · Intern-Decision collection · Editorial description reviewed October 3, 2026
Additional sourceIntern-Decision repository (retrieved October 3, 2026) · Intern-Decision repository · Editorial description reviewed October 3, 2026
Additional sourceIntern-Decision-2B model card (retrieved October 3, 2026; October 4, 2026) · Intern-Decision-2B model card · Editorial description reviewed October 3, 2026
Additional sourceZero-shot IE benchmark, pinned public result snapshot (c45a00c25d2731ac7047def97b91421aaff2f611) (retrieved October 17, 2026)
Additional sourceZero-shot IE benchmark README, pinned commit c45a00c25d2731ac7047def97b91421aaff2f611 (retrieved October 17, 2026)
Additional sourceDuan et al., General Decision Models: Benchmarking and Insights Beyond Jev (arXiv:2610.03935v2) (retrieved October 18, 2026)
Additional sourceSysOne-bench public results.json (retrieved October 23, 2026)
Additional sourceJev Decision Index public index.json (release-v4) (retrieved October 28, 2026)
Additional sourceModel metadata (retrieved October 17, 2026)12
Additional sourceIntern-Decision technical report (Tables 2-6) (retrieved October 17, 2026)
Additional sourceOpenRouter: Popularity index (derived), CC BY 4.0
Additional sourceVercel AI Gateway: Popularity index (derived), CC BY 4.0
Additional sourceOllama: cumulative downloads for the popularity index
Additional sourceReplicate: cumulative runs for the popularity index
Additional sourceGoogle Trending Now: limited feed selection for Germany and the US
Additional sourceCreative Commons Attribution 4.0