whatmodel.sh / decision / Microsoft-Decision-1

Microsoft-Decision-1

+ contribute

Microsoft-Decision-1 is a closed-weights decision model from Microsoft, released 9 October 2026. Licence: proprietary.

Catalogue identity notes

Named hosted endpoint verified 2026-10-31. Perch records the requested model and test date; the service may change its weights. No immutable current-weight equivalence asserted. Not merged with similarly named open-weight community checkpoints.

Model information sources

Release date ↗

63.2±1.3#76 of 355 decision modelsfrom 43 benchmark results across 6 sources

$ how it compares

The decision models ranked either side of Microsoft-Decision-1, out of 357 with a score.

#	model	score	where	RAM	released
74	Clef Cloudflare	63±1	api $0.24/M input tokens +local	~68GB	2026-10
75	linne-fern 1.0 (27B) MISA-AICORE	63±1	model files
76	Microsoft-Decision-1 Microsoft	63±1	api price unknown		2026-10-09
77	Mercury Decide (free endpoint) Inception	63±2	api price unknown
78	Nace.AI Drex v2.0 (35B-A3B) Nace.AI	63±1	model files

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

$ what it's good at

retrieval & classification▁▁▁·▁▁▁▁▃▃█▅ #35 of 258 · 15% of its score · 8 tests

Beats Perplexity Decider v1.1 (27B), H2O Lightning 31B v1.1, AutoTrust-JEV-27B v1.1 here, though they rank higher overall. Trails StartLux-Decision v1.1 2B, GEV-26B-Decide, JEV-35B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
System One Bench: MIRACL pairwise (Arabic)	91.0% ✎	▂···▁··▃▄▂█▇ beats 24/29	easy · avg 84%	▇ 3.4%	typical
System One Bench: MASSIVE intent (Arabic)	89.0% ✎	▄·▁·▁▃▃▃█▄▇▇ beats 22/29	easy · avg 76%	▇ 3.0%	typical
LeanMCP · scifact rank noul	73.0% ✎	▂····▂···▃▅█ beats 5/11	medium · avg 62%	▇ 2.7%	at its level
LeanMCP · nfcorpus rank noul	36.8% ✎	▃▃·····▃▃▅██ beats 4/11	hard · avg 29%	▇ 2.2%	blunt · at its level
System One Bench: MIRACL rerank (Arabic)	90.0% ✎	▁▂·▁▁·▂·▄▄█▇ beats 28/29	easy · avg 79%	▇ 2.0%	typical
System One Bench: MASSIVE scenario (Arabic)	83.5% ✎	▃▁·▄▃▂█▂▁▃·▃ beats 26/29	medium · avg 60%	▇ 1.3%	at its level
System One Bench: AG News (English)	89.0% ✎	▁···▅▁▅▂▆█▂▅ beats 12/29	easy · avg 88%	▇ 0.4%	blunt
System One Bench: Banking77 (English)	87.0% ✎	█▅▆█▅▃▃▃···▂ beats 10/26	very easy · avg 87%	▇ 0.0%	blunt · noisy

language understanding▂▂▂▃▂▅▆▃█▆▄▃ #61 of 256 · 9.2% of its score · 9 tests

Beats Ainvest-Decision-27B v1.0, Turion Jev 31B, MISA AI 2.0 Decision Plus · oak 1.1 (31B) here, though they rank higher overall. Trails Decider 35B-A3B, Sieve 9B v2, Clef-flash here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
System One Bench: BoolQ (English)	88.5% ✎	▁▁▁▃·▁▃▃▇▇█▅ beats 10/29	easy · avg 86%	▇ 2.4%	typical
System One Bench: OSACT4 offensive (Arabic)	88.5% ✎	▁······▁▃█▇▄ beats 11/29	easy · avg 84%	▇ 2.1%	sharp
System One Bench: SST-2 (English)	95.5% ✎	▁·▃▁▂▂▃▅█·▂▁ beats 19/29	easy · avg 92%	▇ 1.4%	typical
System One Bench: SST-5 (English)	62.5% ✎	▃·▁▁▃·▇▄▄▄█▅ beats 29/29	medium · avg 51%	▇ 1.3%	blunt · at its level
System One Bench: XNLI Arabic (Arabic)	72.0% ✎	▄▁▃▁▄▇█▅▃▁·▃ beats 12/29	medium · avg 68%	▇ 0.7%	at its level
System One Bench: XNLI French (French)	82.0% ✎	▂▅·▁▁▅▅█▃▁·▃ beats 16/29	medium · avg 76%	▇ 0.6%	blunt · at its level
System One Bench: Tweet sentiment (Arabic)	68.5% ✎	▁······▁▁▄█▄ beats 16/29	medium · avg 62%	▇ 0.5%	blunt · at its level
System One Bench: XNLI English (English)	89.3% ✎	▃·▁·▃▇█▄█▃▃▁ beats 18/29	easy · avg 85%	▇ 0.2%	blunt · noisy

1 more test

test	result	vs the others in its run	difficulty	counts	why
System One Bench: DAIR Emotion (English)	57.0% ✎	██▃·····▁··▁ beats 12/29	medium · avg 59%	▇ 0.0%	blunt · at its level

knowledge & reasoning▁▂▂▂▂▃▃▅▅▆█▂ #58 of 263 · 45% of its score · 8 tests

Beats OneJev1.6-27B, baselabs elephant-1 (9B), StartLux-Decision v1.1 9B here, though they rank higher overall. Trails Solar Decide, Clef-flash, Clef-omni here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Etsabary · 1342 text decisions	68.6% ✎	▁··▂·▃▅▁▃▂▂█ beats 7/24	medium · avg 47%	▇▇▇▇▇▇ 16%	sharp · at its level
LeanMCP · scienceqa text	93.2% ✎	▄··▄█▄█▄·█▄▄ beats 4/11	very easy · avg 88%	▇▇ 5.6%	sharp · maxed out
LeanMCP · medqa usmle	80.7% ✎	▄▄▄·▄▄▄▄▄·██ beats 4/11	easy · avg 72%	▇▇ 5.2%	typical
CONFLICTS · source conflict types	58.1% ✎	▃▃··▅▃···▃▃█ beats 0/9	medium · avg 45%	▇▇ 4.1%	at its level
MedQA · Freyer et al. (first-valid accuracy)	79.6% ✎	█··▃▃·▃▃▃·▃█ beats 4/11	easy · avg 72%	▇▇ 4.0%	typical
LeanMCP · pubmedqa	68.0% ✎	▂···▂·▂·▄▄▃█ beats 7/11	medium · avg 54%	▇▇ 3.8%	at its level
LeanMCP · medmcqa	75.9% ✎	▃·▃··▃█▃·▅▅▃ beats 8/11	medium · avg 63%	▇ 3.7%	at its level
LeanMCP · mmlu pro	64.7% ✎	▄··▄▄████··▄ beats 4/11	medium · avg 54%	▇ 2.8%	at its level

workflows▁▁▁▂▁▂▂▆█▄▁▁ #8 of 114 · 30% of its score · 18 tests

Beats baselabs mammoth-1 (27B), Nace.AI Drex Max v1 (27B), Perplexity Decider v1.2 (27B) here, though they rank higher overall.

test	result	vs the others in its run	difficulty	counts	why
LeanMCP · mcpbench server routing	94.6% ✎	█·███··▄···▄ beats 4/9	easy · avg 82%	▇▇ 5.3%	sharp · maxed out
LeanMCP · mcpmark server routing	48.8% ✎	▂··▂··▄█▂·▂▄ beats 6/11	medium · avg 36%	▇ 3.6%	at its level
LeanMCP · rewardbench2	74.9% ✎	▄▄▄·▄▄▄▄█·█▄ beats 7/11	medium · avg 58%	▇ 3.2%	at its level
System One Bench: Typed decisions (English)	79.0% ✎	▁▁·▁▁▁▅▄▇██▃ beats 29/29	medium · avg 63%	▇ 3.1%	blunt · at its level
LeanMCP · aegis2 response	82.2% ✎	█··▃·▃·█··▅▅ beats 10/11	easy · avg 70%	▇ 2.8%	typical
LeanMCP · mcpbench tool selection	100.0% ✎	▂··········█ beats 4/9	very easy · avg 89%	▇ 2.3%	sharp · maxed out · noisy
LeanMCP · helpsteer2 helpfulness	0.9012 ordinal levels ✎	▄▄▄▄▄·▄·█▄▄▄ beats 9/10	avg 1.357 ordinal levels	▇ 2.1%	blunt
LeanMCP · sst5	0.5418 ordinal levels ✎	▂·····▂▂▄█▄▂ beats 4/11	avg 0.879 ordinal levels	▇ 2.0%	blunt

10 more tests

test	result	vs the others in its run	difficulty	counts	why
LeanMCP · atbench500	86.8% ✎	▃··▅▅█··▅··▅ beats 8/11	medium · avg 63%	▇ 1.9%	sharp · noisy
LeanMCP · prompt injections	73.3% ✎	▂▂▂▂█·▄▂···▂ beats 5/11	medium · avg 62%	▇ 1.4%	at its level
LeanMCP · mcpmark tool selection	100.0% ✎	▂··········█ beats 4/11	very easy · avg 96%	▇ 1.2%	sharp · maxed out · noisy
Perch semantic lint · private (Youden J)	0.4225 J ✎	█····▄█▄···█ beats 4/7	avg 0.380 J	▇ 0.5%	blunt
Perch semantic lint · public (Youden J)	0.3645 J ✎	█····▄█▄···█ beats 2/7	avg 0.363 J	▇ 0.5%	blunt
Perch security detection · public (Youden J)	0.0563 J ✎	█···▄▄···▄▄█ beats 3/7	avg 0.052 J	▇ 0.1%	blunt
Perch bug detection · private (Youden J)	0.0484 J ✎	▅··▃▃█·····▃ beats 6/7	avg 0.027 J	▇ 0.1%	blunt
Perch bug detection · public (Youden J)	0.0512 J ✎	█▄·█·▄▄····▄ beats 6/7	avg 0.026 J	▇ 0.1%	blunt
Perch security detection · private (Youden J)	0.0142 J ✎	█▃▃··▃·▃···▃ beats 3/7	avg 0.017 J	▇ 0.1%	blunt
LeanMCP · jailbreak classification	96.5% ✎	▂·▄··▂···█▆▂ beats 5/11	very easy · avg 95%	▇ 0.0%	blunt · maxed out · noisy

$ where to run it

Through an API

api · OpenRouter · price unknown · openrouter.ai ↗

Model repository and documentation: openrouter.ai ↗

$ how the score was worked out · LeanMCP · JevBench 52%, System One Bench (Ali Alhashi… 22%, Etsabary · decision benchmarks 16% and 3 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
LeanMCP · JevBench	18	▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 52%	MIT
System One Bench (Ali Alhashimi)	16	▇▇▇▇▇▇▇ 22%	CC-BY-4.0
Etsabary · decision benchmarks	1	▇▇▇▇▇ 16%	MIT
Do We Need an LLM to Know Why Sources Conflict? Decision Models for Knowledge-Conflict Typing in RAG	1	▇ 4%	CC-BY-4.0
Freyer et al. · MedQA	1	▇ 4%	CC BY 4.0
Perch benchmark results	6	▇ 1%	CC BY 4.0
prior · pull toward the middle		▇ 1%
weighted average		= 63.2

$ 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