whatmodel.sh / decision / InternLM Intern-Decision Qwen3.5-4B

InternLM Intern-Decision Qwen3.5-4B

+ contribute

InternLM Intern-Decision Qwen3.5-4B is an open-weights decision model from Shanghai AI Laboratory (InternLM). Its published size is 4.66B parameters. Licence: apache-2.0.

Catalogue identity notes

Exact Decision Index tested engine intern-decision-4b; reported model InternLM Intern-Decision Qwen3.5-4B. Type: full fine-tune; technique: autoregressive. Implementation: https://github.com/internlm/Intern-Decision. Comparison entry; no current run command, API price or speed inferred from the benchmark. Current repository metadata checked separately; an unpinned alias is not evidence that today’s weights reproduce this historical run.

Model information sources

· Weights licence ↗ · Parameter count ↗

Publisher parameter count; includes the served backbone, not just adapter size.

48.1±1.0#220 of 355 decision modelsfrom 66 benchmark results across 5 sources

$ how it compares

The decision models ranked either side of InternLM Intern-Decision Qwen3.5-4B, out of 357 with a score, then more from Shanghai AI Laboratory (InternLM).

#	model	score	where
220	StartLux-Decision v1.1 0.8B StartLux	49±1	model files
221	mmastrac diffgemma (decision readout) vllm-project	48±1	source code
222	InternLM Intern-Decision Qwen3.5-4B Shanghai AI Laboratory (InternLM)	48±1	model files
223	Solar Decide Flash Upstage	48±1	api price unknown
224	razorback16 diffgemma (decision readout) vllm-project	48±1	source code
	more from Shanghai AI Laboratory (InternLM)
280	InternLM Intern-Decision Qwen3.5-2B Shanghai AI Laboratory (InternLM)	37±1	model files
310	InternLM Intern-Decision Qwen3.5-0.8B Shanghai AI Laboratory (InternLM)	30±1	model files

see all decision models · find the ones that fit your machine

$ what it's good at

tools & automation▁▃▁▁▁▂▃▄▆██▆ #144 of 250 · 10% of its score · 9 tests

Beats Cerno 4B v1.1, Trio-Spark DI (9B), Decider 35B-A3B here, though they rank higher overall. Trails JevK5 Lite, Amber 0.4B, Crivo-1B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: BFCL (case exact accuracy) submitted	92.7% ✎	▁▂▁▁▁▂▂▄▅▆█▁ beats 105/246	very easy · avg 92%	▇ 2.1%	sharp · noisy
AgentDecisionBench: L1 BFCL v4 + When2Call (3,991 items)	73.4% ✎	█·····▄·▄▄·▄ beats 2/5	easy · avg 75%	▇ 2.1%	at its level
Decision Index 0.4.1: API-Bank (accuracy) submitted	76.0% ✎	▁▁▁▁▁▁▁▂▃▃█▂ beats 106/236	easy · avg 70%	▇ 1.8%	sharp · at its level · noisy
AgentDecisionBench: L2 DecisionGym common states (200 tasks)	75.2% ✎	██···█·██··█ beats 4/5	medium · avg 57%	▇ 1.6%	sharp · at its level · noisy
Decision Index 0.4.1: When2Call (accuracy) submitted	57.8% ✎	▁▁▁▂▃▅▅▆█▇█▃ beats 62/246	easy · avg 72%	▇ 0.8%	at its level · noisy
Decision Index 0.4.1: Home appliance simulator (case exact accuracy) submitted	12.5% ✎	█▁▁▃▄▄▆▄▁▂▁▃ beats 75/243	medium · avg 37%	▇ 0.7%	sharp · at its level · noisy
Decision Index 0.4.1: ToolRet (nDCG@10) submitted	52.8% ✎	▁▁▁▁▁·▁▁▃▄█▁ beats 35/246	medium · avg 60%	▇ 0.7%	blunt · at its level
AgentDecisionBench: L2 DecisionGym-Shift (100 tasks)	75.9% ✎	█··········█ beats 0/1		—	not counted yet: no other model in the same run to compare it with

1 more test

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: RouterBench (selected quality (quality objective)) submitted	79.8% ✎	no other model yet		—	failed run: Publisher excludes this protocol because its prompt reveals GPT-4 as the best route. Evidence: https://huggingface.co/spaces/multimodalart/jev-decision-index/blob/c130faf65f82e81d3bdefb01e758c712dc5c5199/data/index.json

retrieval & classification▁▁▁·▁▁▁▁▃▃█▅ #175 of 258 · 4.3% of its score · 7 tests

Beats Solar Decide, JevK5 9B v0.3.3, Aplomb 1 here, though they rank higher overall. Trails Mola-68M, Bosun v3.1 1.7B LoRA, Dinah-1 here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: BRIGHT (nDCG@10) submitted	33.1% ✎	▁▁▁▁▁▁▁▂▃▄▇█ beats 62/246	medium · avg 38%	▇ 1.5%	blunt · at its level
Decision Index 0.4.1: Amazon ESCI (macro-F1) submitted	45.3% ✎	▁▁▁▁▁▁▂▂▃▃█▅ beats 83/246	medium · avg 47%	▇ 1.4%	at its level
Decision Index 0.4.1: HoVer (accuracy) submitted	60.5% ✎	▆▆▇▅▄▄▃▇▄█▆▄ beats 52/246	easy · avg 74%	▇ 0.4%	blunt · at its level
Decision Index 0.4.1: PhishNChips phishing decisions (accuracy) submitted	61.5% ✎	█▅▅▄▃▂▁▁▁▁·▁ beats 92/245	easy · avg 70%	▇ 0.4%	at its level · noisy
Decision Index 0.4.1: CLINC150+OOS (macro-F1) submitted	72.7% ✎	▁▁▁▁▁▁▁▂▃▆█▄ beats 45/238	easy · avg 83%	▇ 0.3%	sharp · noisy
Decision Index 0.4.1: BANKING77 (macro-F1) submitted	62.6% ✎	▁·▁▁▁▁▁▁▃█▇█ beats 27/238	easy · avg 79%	▇ 0.2%	noisy
Decision Index 0.4.1: SGD/SGD-X (macro-F1) submitted	66.5% ✎	no other model yet		—	failed run: Publisher reports a builder bug exposing dialogue turns after the classified turn; awaiting corrected rerun. Evidence: https://huggingface.co/spaces/multimodalart/jev-decision-index/blob/c130faf65f82e81d3bdefb01e758c712dc5c5199/data/index.json

language understanding▂▂▂▃▂▅▆▃█▆▄▃ #179 of 256 · 17% of its score · 9 tests

Beats Nari Decision 4B v3.2 here, though they rank higher overall. Trails DecisionTune 1.0, vera-spark, jdi-ptr14-0.8b here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: HellaSwag (accuracy) submitted	77.5% ✎	▂▄▂▂▄▇▃▇█▅▃▁ beats 75/246	very easy · avg 85%	▇▇ 5.1%	sharp
Decision Index 0.4.1: NLI4CT (macro-F1) submitted	77.8% ✎	▁▁▁▂▂▃▃▄▇▇█▄ beats 124/246	easy · avg 74%	▇▇ 4.0%	at its level
Decision Index 0.4.1: ANLI (macro-F1) submitted	58.4% ✎	▁▂▃▅▅█▂····▁ beats 100/246	medium · avg 60%	▇ 3.6%	at its level
Decision Index 0.4.1: iSarcasmEval (Sarcasm F1 · track A, English) submitted	46.4% ✎	▁·▁·▁▁▂▄▅█▇▄ beats 116/246	medium · avg 44%	▇ 1.4%	at its level
Decision Index 0.4.1: ContractNLI (macro-F1) submitted	62.0% ✎	▁▁▁▁▁▁▁▂▃▇▇█ beats 52/246	easy · avg 71%	▇ 0.8%	at its level · noisy
Decision Index 0.4.1: VAST (macro-F1) submitted	40.7% ✎	▁▁▁▂▅▄▄▂▄▆█▄ beats 35/246	medium · avg 62%	▇ 0.7%	at its level
Decision Index 0.4.1: ACOS (per-review F1) submitted	15.9% ✎	▁··▂▃▃▅▇█▇▃▃ beats 83/246	hard · avg 22%	▇ 0.6%	noisy
Decision Index 0.4.1: RAGTruth (F1 on hallucinated class) submitted	71.9% ✎	▁▁▁▁▁▁▁▁▁▃▆█ beats 108/246	easy · avg 66%	▇ 0.6%	at its level · noisy

1 more test

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: FinEntity (macro-F1) submitted	80.9% ✎	▁···▁▁▁▁▃▆█▃ beats 73/246	easy · avg 84%	▇ 0.6%	noisy

arts & human taste▁▇▄▃▅▅█▅▄▂▁▁ #153 of 247 · 6.3% of its score · 6 tests

Beats Jegue 0.1 (9B), Standard One 8B SH (v4), Sieve 4B here, though they rank higher overall. Trails Crivo-32M v2, Crivo-68M, Crivo-150M v2 here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: New Yorker (accuracy) submitted	58.5% ✎	▁▁▂▃▃▃▇▅█▅▃▁ beats 92/246	medium · avg 61%	▇ 2.8%	at its level
Decision Index 0.4.1: Humicroedit (accuracy) submitted	58.6% ✎	▁·▁▃▂▂▅█▄▂▁▁ beats 82/246	medium · avg 60%	▇ 1.2%	blunt · at its level
Decision Index 0.4.1: BPoMP (accuracy) submitted	81.8% ✎	▁█▄▄▅█▆▅▄▁▁▁ beats 95/246	easy · avg 83%	▇ 1.0%	sharp · noisy
Decision Index 0.4.1: cfcolor (accuracy) submitted	58.5% ✎	▁········▁▇█ beats 87/246	medium · avg 60%	▇ 0.7%	blunt · at its level
Decision Index 0.4.1: POP909-CL (accuracy) submitted	7.0% ✎	▄▅▅▃▆▅▃▅▂▅▃█ beats 87/237	hard · avg 21%	▇ 0.4%	sharp · noisy
Decision Index 0.4.1: Habermas Machine (accuracy) submitted	45.1% ✎	▁▂▄█▇▅▁▁▄▄▁▁ beats 137/246	medium · avg 47%	▇ 0.2%	blunt · at its level

knowledge & reasoning▁▂▂▂▂▃▃▅▅▆█▂ #160 of 263 · 33% of its score · 14 tests

Beats StartLux-Decision v1.1 2B, Solomon v1.1, Standard One 3B SH (v4) here, though they rank higher overall. Trails Saracura PT-BR 4B, Ryotide Qwen v0.3.1 (decision readout), Interfaze lev Qwen3.5-4B LoRA here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: MMLU-Pro (accuracy) submitted	46.1% ✎	▄▄▃▃▂▄▅▇██▅▁ beats 101/246	medium · avg 46%	▇▇▇ 6.5%	sharp · at its level
Decision Index 0.4.1: MMLU (accuracy) submitted	71.4% ✎	▁▄▄▃▃▂▅▇▅█▇▁ beats 102/246	easy · avg 70%	▇▇ 5.4%	sharp · at its level
Decision Index 0.4.1: BBH (accuracy) submitted	62.1% ✎	▄▆▄▄█▇▇▄▃▁▁▁ beats 90/246	easy · avg 66%	▇▇ 4.6%	at its level
Decision Index 0.4.1: ARC-Challenge (accuracy) submitted	91.4% ✎	▂▂▂▂▂▂▁▃▄▆█▆ beats 101/246	very easy · avg 90%	▇▇ 4.4%	sharp
Decision Index 0.4.1: ARC-Easy (accuracy) submitted	96.6% ✎	▁▁▂▁▂▂▂▂▃▆█▅ beats 99/246	very easy · avg 96%	▇ 2.6%	sharp · maxed out · noisy
Decision Index 0.4.1: WinoGrande (accuracy) submitted	68.3% ✎	█▅█▅▄█▆▄▃▃▁▁ beats 71/246	easy · avg 78%	▇ 2.4%	typical
Decision Index 0.4.1: GSM8K (accuracy (4- and 10-choice)) submitted	25.6% ✎	▁··▄█▆▆▆▅▅▃▁ beats 59/246	medium · avg 45%	▇ 2.2%	sharp · at its level
Decision Index 0.4.1: GPQA Diamond (accuracy) submitted	41.3% ✎	▂▆▅▅▆█▆▃▁··▁ beats 123/246	medium · avg 39%	▇ 1.9%	blunt · at its level

6 more tests

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: MuSR (accuracy) submitted	53.7% ✎	▁▁▂█▅▄▁▁▁▁▁▁ beats 52/246	medium · avg 60%	▇ 0.8%	blunt · at its level
Decision Index 0.4.1: SATA-Bench (case exact accuracy) submitted	30.8% ✎	▁▁▁▁▁▁▁▂▅█▂▁ beats 156/246	hard · avg 22%	▇ 0.7%	typical
Decision Index 0.4.1: CRUXEval (accuracy) submitted	45.1% ✎	██▄▃▃▄▁▁···▁ beats 88/246	medium · avg 58%	▇ 0.7%	at its level · noisy
Decision Index 0.4.1: CLadder (accuracy) submitted	60.0% ✎	▁███▆▅▃▃▁▁▁▁ beats 64/246	easy · avg 71%	▇ 0.7%	at its level
Decision Index 0.4.1: ChessBench (accuracy) submitted	12.6% ✎	▄▇▄▄▄▆▄█▄▁▁▁ beats 88/246	hard · avg 16%	▇ 0.4%	blunt
Decision Index 0.4.1: HLE (accuracy) submitted	12.8% ✎	▁▁▂▃▆█▇▅▅▃▁▁ beats 143/246	very hard · avg 13%	▇ 0.0%	blunt

vision▁·▃▂▃▃▇▄▄█▄▃ #58 of 79 · 8.9% of its score · 13 tests

Beats QEV-2B, Shisa DE-2 (repeat), djev diffgemma (decision readout) here, though they rank higher overall. Trails RSI-Jev v5.0 VL 3B, Valen-4B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
JevalKit: VisualDecisionBench images	77.9% ✎	▄▄···█▄·█▄█▄ beats 3/10	easy · avg 78%	▇ 1.1%	blunt
Decision Index Vision 0.4 · Mind2Web (chance-corrected points) submitted	65.8800 points ✎	▁·▁▁▃▁▄▄▃▅█▂ beats 21/61	avg 63.423 points	▇ 1.0%	blunt
Decision Index Vision 0.4 · MMMU-Pro vision (chance-corrected points) submitted	29.5300 points ✎	▁▁▅▁▄▁▄█▄▃▇▇ beats 28/61	avg 22.797 points	▇ 0.9%	blunt
Decision Index Vision 0.4 · Winoground (chance-corrected points) submitted	77.0000 points ✎	▁·▁▁▁▁·▂▁▃▅█ beats 23/61	avg 66.987 points	▇ 0.8%	blunt
Decision Index Vision 0.4: private same skills (points)	51.2900 points ✎	▁▁▃▃▁▂▄▆▃▃▅█ beats 22/61	avg 48.067 points	▇ 0.8%	blunt
Decision Index Vision 0.4 · Moderation (Hateful Memes) (chance-corrected points) submitted	26.6300 points ✎	▆▃▃▆▄▄▇██▁▆▆ beats 19/61	avg 28.443 points	▇ 0.7%	blunt
Decision Index Vision 0.4 · RealWorldQA (chance-corrected points) submitted	58.9400 points ✎	▁·▂▂▁▃▃▅▅▃▂█ beats 28/61	avg 53.651 points	▇ 0.7%	blunt
Decision Index Vision 0.4 · CharXiv (chance-corrected points) submitted	57.1400 points ✎	▂▁▁▁▁▃▁▅▅▆██ beats 10/61	avg 62.137 points	▇ 0.6%	blunt

5 more tests

test	result	vs the others in its run	difficulty	counts	why
Decision Index Vision 0.4 · InfographicVQA (chance-corrected points) submitted	83.0600 points ✎	▁·▁▂▁▁▃▁▂▂▅█ beats 15/61	avg 83.859 points	▇ 0.5%	blunt
Decision Index Vision 0.4 · R-Bench-M (chance-corrected points) submitted	19.1500 points ✎	▂▄▇▆█▄▄▇▂··▁ beats 22/61	avg 19.676 points	▇ 0.5%	blunt
Decision Index Vision 0.4 · BLINK (chance-corrected points) submitted	53.7700 points ✎	▁···▁▂▂▂█▄▂▃ beats 32/61	avg 44.994 points	▇ 0.5%	blunt
Decision Index Vision 0.4 · CV-Bench (chance-corrected points) submitted	74.1000 points ✎	▁▁▁▁▂▁▁▁▄█▃▃ beats 28/61	avg 67.662 points	▇ 0.4%	blunt
Decision Index Vision 0.4 · KIE (CORD+FUNSD) (chance-corrected points) submitted	87.9800 points ✎	▁··▁▁▁▁▁▃▄█▁ beats 14/61	avg 85.184 points	▇ 0.3%	blunt

workflows▁▁▁▂▁▂▂▆█▄▁▁ #60 of 114 · 11% of its score · 6 tests

Beats Clef-omni, Clef-flash, Bespoke Nimble 9B v3 here, though they rank higher overall.

test	result	vs the others in its run	difficulty	counts	why
JevalKit: Procedural Typed Decisions	61.0% ✎	█·███▄·▄▄▄·▄ beats 2/12	easy · avg 67%	▇▇ 4.6%	sharp · at its level
Privacy policy · diagnostic43	65.1% ✎	█········▄·▄ beats 3/3	easy · avg 70%	▇ 2.8%	sharp · at its level
JevalKit: DecisionBench standard eval	60.5% ✎	▃▃·▃···▅█▅·█ beats 2/12	easy · avg 70%	▇ 1.9%	at its level
JevalKit: JevBench public 231	86.6% ✎	█▄▄▄▄▄██▄··▄ beats 7/12	easy · avg 82%	▇ 1.1%	typical
Privacy policy · challenge56	100.0% ✎	▄·······▄··█ beats 3/3	very easy · avg 97%	▇ 0.2%	sharp · maxed out · noisy
JevalKit: Typed Decisions held-out modal labels	80.5% ✎	▃▅··▃·▃▃·█▃█ beats 10/12	easy · avg 74%	▇ 0.1%	blunt · at its level

other tests5 not in a task · 8.8% of its score

test	result	vs the others in its run	difficulty	counts	why
Decision Index 0.4.1: private same skills (equated points)	38.5000 points ✎	▂▂▂▃▃▃▄▅▅▇█▃ beats 90/247	avg 41.206 points	▇ 3.4%	blunt
JevalKit: JevAlt-Bench EN/TR/DE	89.0% ✎	▅··▃▃▅▃█··▅▃ beats 6/12	very easy · avg 87%	▇ 2.7%	typical
Decision Index 0.4.1: private new domains (equated points)	38.9700 points ✎	▃█▃▃▂▆▃▃▄▄▄▄ beats 119/247	avg 38.861 points	▇ 1.9%	blunt
JevalKit: Decision Index public 0.3 (official weighted points)	25.5100 points ✎	▃▃·▃·▃▃▃▅▅·█ beats 2/12	avg 39.649 points	▇ 0.8%	blunt
JevalKit: Chinese decision validation v0.2	91.6% ✎	▃····█·█▃▃█▃ beats 12/12	very easy · avg 88%	▇ 0.1%	blunt

$ where to run it

Download model files ↗
Published model files verified. This is a verified repository snapshot; the benchmark does not establish an identical weight revision. A local run command and hardware requirements have not been verified for this entry. Source · checked 2026-10-31.

Model repository and documentation: huggingface.co/internlm ↗

$ how the score was worked out · Decision Index 0.4.1 73%, JevalKit (Valen Team) 12%, Decision Index Vision 0.4 8% and 2 more

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

source	results	share of the score	licence
Decision Index 0.4.1 · submitted; independent execution unverified	42	▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 73%	CC BY 4.0
JevalKit (Valen Team)	8	▇▇▇▇ 12%	Apache-2.0
Decision Index Vision 0.4 · submitted; independent execution unverified	12	▇▇ 8%	CC BY 4.0
AgentDecisionBench	2	▇ 4%	CC-BY-4.0
FloorIsAwake · Privacy Decision Benchmark	2	▇ 3%	MIT
prior · pull toward the middle		▇ 0%
weighted average		= 48.1

not counted (3)

Decision Index 0.4.1	Decision Index 0.4.1: RouterBench (selected quality (quality objective)) [correct]	79.8%	failed run: Publisher excludes this protocol because its prompt reveals GPT-4 as the best route. Evidence: https://huggingface.co/spaces/multimodalart/jev-decision-index/blob/c130faf65f82e81d3bdefb01e758c712dc5c5199/data/index.json
Decision Index 0.4.1	Decision Index 0.4.1: SGD/SGD-X (macro-F1) [correct]	66.5%	failed run: Publisher reports a builder bug exposing dialogue turns after the classified turn; awaiting corrected rerun. Evidence: https://huggingface.co/spaces/multimodalart/jev-decision-index/blob/c130faf65f82e81d3bdefb01e758c712dc5c5199/data/index.json
AgentDecisionBench	AgentDecisionBench: L2 DecisionGym-Shift (100 tasks) [correct]	75.9%	fewer than three models have this test yet, or this source ran it for this model alone

$ improve this page

Update model information Add a place to run it Add a test result Propose a benchmark Correct a benchmark score Edit data on GitHub

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated 2026-11-03 02:37 UTC · sources and licences · json