PipelineScore
← Back to leaderboard
microsoft

Phi 3.5 MoE 42B

Released 2024-08-20Context 131.072Kphi-3-5-moe
PipelineScore
69.1FEEDER
Ranked #49 of 52 models · 6th percentileRAG is the headline (73.3); throughput is the soft spot (63.2). Best-fit profile: Agentic.

Category breakdown

Score per category, normalized 0–100 against the v1 anchor.

Code
71.5
Reason
70.9
Tool Use
71.3
RAG
73.3
Speed
63.2

Strengths

RAG73.3
Code71.5
Tool Use71.3

Same model, different rigs

Every submission of Phi 3.5 MoE 42B on the 0–100 scale. The spread is the point: where it runs changes what you get.

0255075100

Best 72.3 on rtx-3090-24gb · lowest 66.0 on m3-max-64gb · spread 6.4 pts across 4 runs. Hover a dot for its rig.

Sample tasks

A taste of what the test pack measures. Full prompts are private and rotated daily.

CodeDifficulty 1code-fib-1

Fibonacci function

Write a Python `fib(n)` returning the nth Fibonacci number, O(n).

ReasonDifficulty 1reason-math-1

Train meeting time

Two trains, opposite directions, given speeds and start times — when do they meet?

RAGDifficulty 2rag-extract-1

Extract metrics to JSON

From the context, extract net sales, operating margin, and free cash flow as a JSON object. Numbers only.

Tool UseDifficulty 2tool-schema-1

OpenAPI param selection

Given an OpenAPI schema with limit/offset/sort, fill JSON for 'next 50, recent first.'

RAGDifficulty 2rag-grounding-1

Refuses to fabricate

Context lacks the answer — does the model fabricate or correctly say it can't?

Compare with

Phi 3.5 MoE 42B vs openai/gpt-oss-20bPhi 3.5 MoE 42B vs DeepSeek R1 671B-A37BPhi 3.5 MoE 42B vs Qwen 3 235B-A22B MoEPhi 3.5 MoE 42B vs DeepSeek V3 671B-A37BPhi 3.5 MoE 42B vs DeepSeek Coder V2 236B