The Known Good Updated 29 Jul 2026
Microsoft ·Open weights (unrestricted) ·Released December 2024·Tracked since 25 Jul 2026

Microsoft: Phi 4 — quality, speed & price analysis

Compare with another model Provider benchmarks Data sources
Quality
#51 of 62
41based on 4 of 7 evaluations
Known Good Index
Speed
#121 of 317
58.9
Output tokens per second
Price
#43 of 370
$0.09
Blended per 1M tokens
In $0.07 · Out $0.14
Cache hit price
unranked
per 1M tokens
Context
#364 of 375
16k
tokens (~25 pages)
Summary

Microsoft: Phi 4 is in the lowest quarter of the models we track for quality, and inexpensive for its standing, with mid-range throughput. It accepts text input, outputs text, and supports a 16k-token context window.

It scores 41 on the Known Good Index, well below the tracked average of 64, from 4 of the seven component evaluations. Measured against the tracked average, its strongest result is MATH Level 5 (65 against 55) and its weakest is Mock AIME 2024–25 (14 against 53).

Pricing is $0.07 per 1M input tokens (below the $2.10 average) and $0.14 per 1M output (below the $9.62 average). At 59 tokens per second it sits in the upper middle of the tracked range.

Generated from tracked data on 29 Jul 2026 · how these comparisons are computed →

Technical specifications
ReasoningNo
Input modalitytext
Output modalitytext
Context window16,384 tokens
Max output16,384 tokens
WeightsOpen weights · Open weights (unrestricted)
Knowledge cutoffJune 2024
Tool callingNo
Providers1 host · compare →
Highlights — where Microsoft: Phi 4 sits
Jump to section

Quality

How Microsoft: Phi 4 compares against every tracked model

Known Good IndexCapability indices
Known Good Index

Composite of GPQA Diamond, Mock AIME 2024–25, MATH Level 5, SWE-bench Verified, LiveBench, Humanity's Last Exam and Terminal-Bench, normalised 0–100. Scores are ingested from Epoch AI, LiveBench; we do not run these evaluations.

💡 A lightbulb marks a model its provider publishes as a reasoning model
The Known Good

Quality Breakdown

Microsoft: Phi 4 on each evaluation feeding the index, against the tracked average

Evaluation scores

Bar = Microsoft: Phi 4 · grey marker = average across tracked models · scores ingested from Epoch AI, LiveBench

GPQA Diamond56 / 63
Mock AIME 2024–2514 / 53
MATH Level 565 / 55
LiveBench42 / 52
The Known Good

Arena Elo

Human preference rating from blind pairwise votes

Text Arena

Elo from blind pairwise votes, ingested from Arena

1217
Elo · 95% CI ±5
#279
of 386 ranked models
24k
votes counted
1256
style-controlled (#285)

Ingested 29 Jul 2026.

The Known Good

Price & Cost

Published API pricing by token type

BlendedInput vs outputQuality per dollar
Blended Price

USD per 1M tokens at a 3:1 input:output blend. Live from OpenRouter, cross-checked against models.dev. Lower is better.

The Known Good
We publish blended price rather than cost-per-task. Cost-per-task requires token counts from an evaluation suite we do not run. Blended price is computed from published rates and is fully reproducible — see the pricing methodology.

Context Window

Maximum input tokens accepted

Context windowMax output tokens
Context Window

Maximum input context from provider metadata via models.dev and OpenRouter. Higher is better.

The Known Good

Speed

Median output throughput across hosting providers

Output speedBy provider

Latency

Time to first token, median across providers

Latency (TTFT)

Seconds to first streamed chunk, median across tracked providers. Derived from OpenRouter endpoint telemetry. Lower is better.

The Known Good

Providers

Hosts serving Microsoft: Phi 4, with their own price and speed

1 provider
Full provider benchmarks →
ProviderInput $/MOutput $/M BlendedSpeedTTFTContext
DeepInfra $0.07 $0.14 $0.09 57 0.2s 16k

Per-host pricing and telemetry from OpenRouter. Speed is output tokens per second; TTFT is time to first token.

The Known Good

History

How price and quality have moved since we began tracking this model

Quality & price over time

Since 25 Jul 2026

Known Good Index 41 · based on 4 of 7 evaluations
Blended price $0.09 per 1M · observed 29 Jul 2026
Output speed 59 tok/s · observed 29 Jul 2026
Time to first token 0.2s

No recorded changes for this model since tracking began.

The Known Good

Comparisons

Head-to-head pages pairing Microsoft: Phi 4 with other tracked models

Head-to-head pages are generated for the top 50 models by Known Good Index, which Microsoft: Phi 4 is not currently in. Its nearest tracked models by index are below.

claude-3-5-sonnet-20241022 grok-2-1212 mistral-large-2411 OpenAI: GPT-4.1 Google: Gemma 3 27B gemma-3-27b-it Meta: Llama 3.3 70B Instruct claude-opus-4-1-20250805
Sources for this page
Pricing & providers — OpenRouter Metadata — models.dev Benchmarks — Epoch AI (CC-BY 4.0)Benchmarks — LiveBench (Apache-2.0)Elo — Arena Full attribution →