The LLM benchmark aggregator.

Raw Phi-4 mini direct logits

Microsoft

Strongest in Decisions (#48 of 85), weakest in Calibration (#57 of 85). Above par in 1 of 3 scopes. Among the models it meets almost everywhere, it finishes behind DeepSeek V4.1 Flash and GPT-6 Luna and ahead of openJev Verdict 1.4 and System One Open.

§ 1 · Profile

What it is good at

Bars run from 50 — the average of the models each source lists — so right of the line is above par. The middle column is the gap to whoever leads that scope.

Decisions49.2−24.1#48/851/1

Routing & classification50.5−23.5#49/851/1

Calibration48−24.7#57/851/1

§ 2 · Head to head

What it beats, and what beats it

The same models turn up scope after scope. Counted once: where both were placed, who finished higher. Bars run right for Raw Phi-4 mini direct logits, left for the other.

Laya 322Mwon 2 · lost 0

Djev Sparkwon 2 · lost 0

Mirrorwon 3 · lost 0

Raw Qwen3 0.6B direct logitswon 3 · lost 0

GLiNER2won 3 · lost 0

GLiNER2 largewon 3 · lost 0

Raw Qwen3 1.7B direct logitswon 3 · lost 0

CLM 8Bwon 3 · lost 0

SimpleJevwon 3 · lost 0

GLiNER2.5 smallwon 3 · lost 0

openJev Verdictwon 3 · lost 0

GLiNER2.5 multiwon 3 · lost 0

kev 0.5Bwon 3 · lost 0

Decider 2Bwon 3 · lost 0

kev 8Bwon 3 · lost 0

JevActwon 3 · lost 0

kev 0.6Bwon 3 · lost 0

OpenDecisionwon 3 · lost 0

Malkuth 2Bwon 3 · lost 0

Open Alternative Jevwon 3 · lost 0

Raw Qwen3 8B direct logitswon 2 · lost 1

Open Jev JSON Canvaswon 2 · lost 1

Raw Qwen3 4B Instruct direct logitswon 2 · lost 1

System Onewon 2 · lost 1

kev 4Bwon 2 · lost 1

Verdict Smallwon 2 · lost 1

Smalljev semantic-v9won 2 · lost 1

Alibaba GTE Reranker ModernBERT-basewon 2 · lost 1

Certo v1won 2 · lost 1

Mixedbread mxbai-rerank-base-v2won 2 · lost 1

BAAI bge-reranker-v2-m3won 2 · lost 1

Open Jev Deberta V3 Largewon 2 · lost 1

Spark S1 4B V6won 2 · lost 1

Qwen3.5-0.8B Decision Modelwon 2 · lost 1

Typecastlmwon 2 · lost 1

Open-Jev 2Bwon 2 · lost 1

Layawon 2 · lost 1

System One Openwon 2 · lost 1

openJev Verdict 1.4won 2 · lost 1

Decision Fastwon 1 · lost 2

OpenSourceJevwon 1 · lost 2

Bespoke Nimble 9Bwon 1 · lost 2

Jeffwon 1 · lost 2

Lev 350Mwon 1 · lost 2

Vonwon 1 · lost 2

Decision Machine 1won 1 · lost 2

Decision 2Bwon 1 · lost 2

OpenJevwon 1 · lost 2

Gemma 4 31Bwon 0 · lost 2

Malkuth 4Bwon 0 · lost 3

SimpleJev Qwen3.6-35B-A3Bwon 0 · lost 3

Open-Jev 9Bwon 0 · lost 3

Qwen3.5-9B Jev-like data-mix v2won 0 · lost 3

Jev Localwon 0 · lost 3

Qwen3 Reranker 4Bwon 0 · lost 3

Jobe Qwen3.5-4Bwon 0 · lost 3

Jev Omniwon 0 · lost 3

SemIfwon 0 · lost 3

Metask Jev 4Bwon 0 · lost 3

Standard One 8Bwon 0 · lost 3

Decider 35B A3Bwon 0 · lost 3

Winnow-12B Q8won 0 · lost 3

JEV Qwen3.5-9B Base NVFP4won 0 · lost 3

Gemini 3.1 Flash Litewon 0 · lost 3

ZeroEntropy zerank-2won 0 · lost 3

local-jev Qwen3.5-4Bwon 0 · lost 3

Jqvwon 0 · lost 3

reflex 4Bwon 0 · lost 3

Openjev Sglangwon 0 · lost 3

LitJevwon 0 · lost 3

Reflex 27Bwon 0 · lost 3

JevK5 v0.2.0won 0 · lost 3

swanOnewon 0 · lost 3

Decider-4b v2won 0 · lost 3

Cygnetwon 0 · lost 3

JevOnewon 0 · lost 3

SimpleJev Qwen3.8-27Bwon 0 · lost 3

Hopperwon 0 · lost 3

NInfer Qwen3.8-27B NVFP4won 0 · lost 3

NInfer Qwen3.8-Flash-Next mixedwon 0 · lost 3

Jev 1.13.0won 0 · lost 3

Instinctwon 0 · lost 3

Gemma 4 31B IT (Autoloops)won 0 · lost 3

Djevwon 0 · lost 3

GPT-5.6 Lunawon 0 · lost 3

GPT-6 Lunawon 0 · lost 3

DeepSeek V4.1 Flashwon 0 · lost 3

§ 3 · Sources

Where the numbers come from

1 publication, 2 figures. Every one links to the page it was read from.

JevBench ↗2 measuresread 2026-09-26

JevBench · Intelligence 41.8%JevBench · Calibration 58.8%

Badge

[![PublicAI Index](https://publicai.io/model-index/badge?model=raw-phi-4-mini-direct-logits)](https://publicai.io/model-index/m/raw-phi-4-mini-direct-logits)