Skip to content

Microsoft · Chat

Phi 4

Microsoft 🇺🇸ChatPhi familyOpen weights. One of its sources says the weights are publicly available.The maker currently plans to deprecate it within the next 180 days (17 Jan 2027). Plans can change.
From Aihubmix's listing:
Phi-4 is a state-of-the-art open model based on a combination of synthetic datasets, curated public domain website data, and acquired academic books and QA datasets. The approach aims to ensure that small, efficient models are trained using data focused on high quality and advanced reasoning.
llminfo IQ
105.9
Market median price
$0.0875 blended / 1M
Max context
16K
Listings
14

Evidence and sources

🤗 microsoft/phi-4llminfo IQ 105.9, ranked among chat models. From 144 benchmarks. Plenty of data behind it.How llminfo IQ is computedCoding IQ 77.6, 95% interval 66–89. From 4 benchmark families across 4 sources. Little data behind it, so treat it as a rough estimate. Uses the model's best published results. It has its own scale, separate from llminfo IQ.How Coding IQ is computedFair value (51.2). Costs per task about what its benchmark results predict. From 1 benchmark with measured task costs. Little data behind it, so treat it as a rough estimate.How Value for money worksReleased 12 Dec 2024 Training-data cutoff, as its sources state it Maker HQ: US (source)
All rankings: #357 on OpenRouter this week #377 on OpenRouter today #377 on OpenRouter this monthSources: LMArena data, DeepInfra data, LiteLLM data, Models.dev data, OpenRouter data docs, AIHubMix data docs, llminfo researchPrimary data from LMArena from DeepInfra: name from LiteLLM: deprecation date from Models.dev: weights availability from OpenRouter: knowledge cutoff from AIHubMix: description from llminfo research: release date

Advantages and trade-offs

Compared with 10 chat models, picked mainly by popularity and IQ

Advantages

Open weights available

The weights are public. 3 of the 7 peers that say either way have open weights.

Trade-offs

Deprecation scheduled

The maker plans to deprecate it on 17 Jan 2027, within the next 180 days.

We only make a claim when there's enough data on models of the same type. Updated 22 Nov, 04:34 UTC.Prices come from the model's maker where they sell it, otherwise their official partner, otherwise the market median. This comparison uses 6 maker's prices; 4 market median prices.

Input & output

as stated by the model's providers
Text
Text

Capabilities

Supported operations and provider-stated features

Function callingStructured output

Listings

 Microsoft Azurephi-4Sources: GenAI Prices data, LiteLLM data, Models.dev data16K$0.07$0.14——
 Aihubmixaihub-Phi-4Source: AIHubMix data docs16K$0.12$0.48——
 Fastroutermicrosoft/phi-4Source: FastRouter data16K$0.07$0.14——
 Flowdotmicrosoft/phi-4Source: FlowDot data16K$0.065$0.14——
 Hugging FaceInference by DeepInframicrosoft/phi-4Source: Hugging Face Inference Providers data16K$0.07$0.14——
 Kilomicrosoft/phi-4Sources: Kilo AI Gateway data, Models.dev data16K$0.056$0.112——
 OpenRouterInference by DeepInframicrosoft/phi-4 · bf16/flex · bf16Source: OpenRouter data docs16K$0.056$0.112103477ms
 OpenRouterInference by DeepInframicrosoft/phi-4 · bf16Source: OpenRouter data docs4K$0.07$0.14102319ms
 OpenRoutermicrosoft/phi-4Sources: GenAI Prices data, Models.dev data, OpenRouter data docs16K$0.07$0.14——
 RequestyInference by DeepInfradeepinfra/microsoft/phi-4Source: Requesty data16K$0.07$0.14——
 DeepInframicrosoft/phi-4Sources: DeepInfra data, LiteLLM data16K$0.07$0.14——

3 listings without a price are hidden.

Benchmarks

Named effort levels run from lowest to highest. Select a result to see its source details.

Knowledge

Math

LiveBenchSource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
MATHSource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
MGSMSource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
PhiBenchSource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
U-MATH7 related benchmarks · 21 results
algebra accuracy
Source: U-MATH and mu-MATH
Task set · algebra · gpt-4o-2024-08-06
differential_calc accuracy
Source: U-MATH and mu-MATH
Task set · differential_calc · gpt-4o-2024-08-06
integral_calc accuracy
Source: U-MATH and mu-MATH
Task set · integral_calc · gpt-4o-2024-08-06
multivariable_calculus accuracy
Source: U-MATH and mu-MATH
Task set · multivariable_calculus · gpt-4o-2024-08-06
overall accuracy
Source: U-MATH and mu-MATH
Task set · u_math · gpt-4o-2024-08-06
precalculus_review accuracy
Source: U-MATH and mu-MATH
Task set · precalculus_review · gpt-4o-2024-08-06
Show 3 more resultsHide 3 results
sequences_series accuracy
Source: U-MATH and mu-MATH
Task set · sequences_series · gpt-4o-2024-08-06

Human preference

Arena27 related benchmarks · 27 results
ELO
Source: LMArena data · Nov 2026
ELO - Chinese
Source: LMArena data · Nov 2026
ELO - Creative Writing
Source: LMArena data · Nov 2026
ELO - English
Source: LMArena data · Nov 2026
ELO - Exclude Ties
Source: LMArena data · Nov 2026
ELO - Expert
Source: LMArena data · Nov 2026
Show 21 more resultsHide 21 results
ELO - Japanese
Source: LMArena data · Nov 2026
ELO - Korean
Source: LMArena data · Nov 2026
ELO - Longer Query
Source: LMArena data · Nov 2026
ELO - Multi Turn
Source: LMArena data · Nov 2026
ELO - Non English
Source: LMArena data · Nov 2026
ELO - Russian
Source: LMArena data · Nov 2026
ELO - Spanish
Source: LMArena data · Nov 2026
ELO Style-Controlled
Source: LMArena data · Nov 2026
Hard
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim

General

Factuality

SimpleQASource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim

Coding

HumanEvalSource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim

Agentic

Reasoning

DROPSource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim

Multilingual

Science

GPQASource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim

Instruction following

IFEvalSource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim

Other

Data credits: LMArena, CC BY 4.0 (licence) · Data by LLM Stats · Epoch AI Benchmarking Hub, CC BY 4.0 (licence) · EQ-Bench (Samuel J. Paech) — cite paech2023eqbench · FaithJudge, Vectara; Tamber et al., EMNLP 2025 Industry. Public numeric README leaderboard; results licence not separately established. · Speakleash community evaluation of INCLUDE-base-44 (Romanou et al., arXiv:2411.19799). Public results.csv; no results licence stated. · Hugging Face official benchmark leaderboards; licence varies by benchmark · LLM Chess by Maxim Saplin, Apache-2.0 (Copyright [2024] [Maxim Saplin]) (licence) · Open Arabic LLM Leaderboard (OALL), Apache-2.0 (licence) · pfgen-bench, Preferred Networks. Apache-2.0 except LLM-generated outputs; aggregate scores only. (licence) · SnakeBench (Greg Kamradt / ARC Prize Foundation), MIT (licence) · Toloka, U-MATH and mu-MATH public leaderboard. MIT Space; generated solution outputs retain underlying model terms and are not ingested. (licence) · UzLiB, Tahrirchi (Abror Shopulatov). Public LEADERBOARD.md numbers; repository MIT covers software and documentation, no separate results licence stated and none asserted. · Vectara Hallucination Leaderboard, Apache-2.0 (licence)

Also known as: aihub-phi-4, azure_ai/phi-4, deepinfra/microsoft/phi-4, microsoft/phi-4

Like this one. Tap or hover any dotted value, label or icon to see where it comes from and how it is measured. open details when tapped or hovered. How this data works