Skip to content
Model Pareto

Phi-4

Microsoft AIOpen weightsartificialanalysis.ai ↗
Compare with nearest rivalsCompare with…
Input
$0.125/M tok
Output
$0.50/M tok
Blended
$0.219/M tok · 3:1
Speed
40tok/s
Context
16Ktokens

Long-tail import from Artificial Analysis (2026-09-24); AA variant: Phi-4. Price = AA median host price.

Rankings

Self-hostingEstimated

How we estimate →

What it would cost to run these open weights yourself on rented GPUs, compared with the API price above.

Hardware
1× B200 180GB
BF16 weights
Throughput
~7,374 tok/s
many requests batched
Price
$0.12–1.08 /M tok
busy → light use
On your own machine
RTX 5090 32GB (int4, ~170 tok/s single-stream)
single consumer GPU or Mac
Assumptions (5)
  • BF16 weights (14.7B params) + 25% KV-cache headroom ≈ 37 GB
  • 1× B200 180GB at $5.98–$14.24/GPU-hour on-demand (2026-09-24)
  • ~7,374 output tok/s aggregate at batch 211 (bandwidth-bound)
  • Blended 3:1 input:output; prefill ~36,161 tok/s
  • Low = 75% utilization at the low GPU price; high = 20% at the high price

Benchmark scores

~ italic, dashed = no published score yet, estimated from related benchmarks.

Writing & chat

MMMLUEstimated
~last of 5 (estimated)below measured range (<81.3%)
  1. #1Gemini 3.1 Pro92.6%
  2. #2Gemma 4 31B88.4%
  3. #4gpt-oss-120b81.3%
  4. ~#5Phi-4Estimated<81.3%
No published score — estimated from related benchmarks
LMArena Text, Non-English (Elo)Verified
#52 of 531233 Elo
  1. #1Claude Opus 5.51501
  2. #2Claude Fable 5.1Few votes1496
  3. #51NVIDIA Nemotron 3 Nano 30B A3B1286
  4. #52Phi-41233
  5. #53Olmo 3.1 32B Think1231
lmarena.ai · 2026-09-13
LMArena Text (Elo)Verified
#53 of 531256 Elo
  1. #1Claude Opus 5.51509
  2. #2Claude Fable 5.11501
  3. #52Olmo 3.1 32B Think1286
  4. #53Phi-41256
lmarena.ai · 2026-09-13
EQ-Bench Creative Writing v3 (Elo)Estimated
~#24 of 26 (estimated)~1147 Elo
  1. #1GPT-6 Astra2173
  2. #2Claude Fable 5.12162
  3. #23Gemma 4 31B1368
  4. ~#24Phi-4Estimated1147
  5. #24gpt-oss-120b961
No published score — estimated from related benchmarks

Hard reasoning

AIME (latest)Estimated
~#14 of 25 (estimated)~38.6%
  1. #1Inkling97.1%
  2. #2Muse Glimmer94.7%
  3. #13Phi-4-reasoning-plus78.0%
  4. ~#14Phi-4Estimated38.6%
  5. #14Mistral Large 338.0%
No published score — estimated from related benchmarks
AA Intelligence IndexEstimated
~#73 of 75 (estimated)~4.8
  1. #1Claude Opus 5.557.6
  2. #2Claude Sonnet 5.556.0
  3. #72Ministral 3 8B5.5
  4. ~#73Phi-4Estimated4.8
  5. #73Ministral 3 3B4.8
No published score — estimated from related benchmarks

Agents

OSWorld-VerifiedEstimated
~#11 of 12 (estimated)~62.4%
  1. #1Kimi K384.8%
  2. #2Qwen3.8 27B84.3%
  3. #10Muse Glimmer65.9%
  4. ~#11Phi-4Estimated62.4%
  5. #11Claude Haiku 4.550.7%
No published score — estimated from related benchmarks
OSWorld 2.0Estimated
~#3 of 5 (estimated)~53.8%
  1. #1GLM-5.3 Flash59.1%
  2. #2Kimi K358.3%
  3. ~#3Phi-4Estimated53.8%
  4. #3GPT-5.6 Terra50.2%
No published score — estimated from related benchmarks
BrowseCompEstimated
~#17 of 18 (estimated)~33.0%
  1. #1Atria Dawn Preview92.5%
  2. #2GPT-6 Astra91.5%
  3. #16Nemotron 3 Ultra44.4%
  4. ~#17Phi-4Estimated33.0%
  5. #17Mistral Small 421.3%
No published score — estimated from related benchmarks
GDPval-AA (Elo)Estimated
~#59 of 87 (estimated)~581 Elo
  1. #1Claude Opus 5.51846
  2. #2Claude Fable 5.11735
  3. #58gpt-oss-120b596
  4. ~#59Phi-4Estimated581
  5. #59Granite 4.2 30B564
No published score — estimated from related benchmarks
AutomationBench-AAEstimated
~#80 of 80 (estimated)~0.2%
  1. #1Claude Opus 5.569.5%
  2. #2DeepSeek V4.1 Flash68.9%
  3. #74Magistral Small 1.20.2%
  4. ~#80Phi-4Estimated0.2%
No published score — estimated from related benchmarks
τ-Bench Banking (AA)Estimated
~last of 89 (estimated)below measured range (<3.3%)
  1. #1Muse Spark 1.350.5%
  2. #2GLM-5.350.3%
  3. #88Llama 4 Scout3.3%
  4. ~#89Phi-4Estimated<3.3%
No published score — estimated from related benchmarks