Skip to content
Model Pareto

Phi-4-reasoning-plus

Microsoft AIOpen weightshuggingface.co ↗
Input
—
Output
—
Speed
~47tok/sest.
Context
33Ktokens

14B dense reasoning model (Phi-4 + SFT + RL). No API price found (not on OpenRouter).

Rankings

Self-hostingEstimated

How we estimate →

What it would cost to run these open weights yourself on rented GPUs. No API sells this model, so this is its price on the chart.

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.

Coding

DeepSWE v1.1Estimated
~#43 of 46 (estimated)~31.8%
  1. #1Gemini 4 Argon78.8%
  2. #2DeepSeek V4.1 Flash 101777.0%
  3. #42Nex-N2.5-Mini36.1%
  4. ~#43Phi-4-reasoning-plusEstimated31.8%
  5. #43Hy328.0%
No published score — estimated from related benchmarks
SciCodeEstimated
~#50 of 119 (estimated)~48.1%
  1. #1Claude Fable 5.267.7%
  2. #2Claude Opus 5.566.9%
  3. #49Quasar 438B48.1%
  4. ~#50Phi-4-reasoning-plusEstimated48.1%
  5. #50Solar Open2 250B48.0%
No published score — estimated from related benchmarks
~#15 of 26 (estimated)~46.7
  1. #1Claude Opus 5.561.7
  2. #2Claude Opus 559.7
  3. #14Grok 4.647.0
  4. ~#15Phi-4-reasoning-plusEstimated46.7
  5. #15Grok 4.746.6
No published score — estimated from related benchmarks

Hard reasoning