Phi-4-reasoning-plus
- 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
SWE-bench VerifiedEstimated
~#9 of 21 (estimated)~77.3%
- #1Claude Opus 596.0%
- #2Claude Sonnet 585.2%
- #7Mistral Medium 3.577.6%
- ~#9Phi-4-reasoning-plusEstimated77.3%
- #9Ripple-65B Preview76.5%
No published score — estimated from related benchmarks
SWE-bench Pro (Public)Estimated
~#11 of 24 (estimated)~61.6%
- #1Claude Fable 5.293.0%
- #2Claude Opus 5.589.9%
- #10Claude Sonnet 563.2%
- ~#11Phi-4-reasoning-plusEstimated61.6%
- #11Nex-N2.5-Pro61.2%
No published score — estimated from related benchmarks
DeepSWE v1.1Estimated
~#43 of 46 (estimated)~31.8%
- #1Gemini 4 Argon78.8%
- #2DeepSeek V4.1 Flash 101777.0%
- #42Nex-N2.5-Mini36.1%
- ~#43Phi-4-reasoning-plusEstimated31.8%
- #43Hy328.0%
No published score — estimated from related benchmarks
Terminal-Bench 4.0Estimated
~#28 of 102 (estimated)~25.6%
- #1Kimi K3.1 Pro67.2%
- #2Claude Sonnet 5.563.6%
- #26MiniMax-M3.1-Flash25.8%
- ~#28Phi-4-reasoning-plusEstimated25.6%
- #28Qwen3.8 Flash-Next25.3%
No published score — estimated from related benchmarks
LiveCodeBenchLab-reported
#18 of 2753.1%
- #1Qwen3.8-Omni-Flash92.6%
- #2Qwen3.8 Flash-Next91.9%
- #17GLM-5.3 NVFP453.5%
- #18Phi-4-reasoning-plus53.1%
- #19Falcon-H1-34B-Instruct49.7%
huggingface.co · 2025-04-30
SciCodeEstimated
~#50 of 119 (estimated)~48.1%
- #1Claude Fable 5.267.7%
- #2Claude Opus 5.566.9%
- #49Quasar 438B48.1%
- ~#50Phi-4-reasoning-plusEstimated48.1%
- #50Solar Open2 250B48.0%
No published score — estimated from related benchmarks
AA Coding Agent IndexEstimated
~#15 of 26 (estimated)~46.7
- #1Claude Opus 5.561.7
- #2Claude Opus 559.7
- #14Grok 4.647.0
- ~#15Phi-4-reasoning-plusEstimated46.7
- #15Grok 4.746.6
No published score — estimated from related benchmarks
Hard reasoning
GPQA DiamondLab-reported
#116 of 19268.9%
- #1GPT-6 Astra96.1%
- #2Gemini 3.8 Flash95.3%
- #115Step3 VL 10B69.0%
- #116Phi-4-reasoning-plus68.9%
- #117gpt-oss-20b68.8%
huggingface.co · 2025-04-30
Humanity's Last ExamEstimated
~#88 of 199 (estimated)~17.6%
- #1Claude Fable 5.274.3%
- #2Claude Opus 5.561.4%
- #87K2 Horizon 7B18.2%
- ~#88Phi-4-reasoning-plusEstimated17.6%
- #88G9v3-39A5B17.5%
No published score — estimated from related benchmarks
AIME (latest)Lab-reported
#22 of 3378.0%
- #1GLM-5.3 NVFP499.5%
- #2Beam 1.0 501B A23B97.7%
- #21Magistral Small 1.280.3%
- #22Phi-4-reasoning-plus78.0%
- #23Mistral Large 338.0%
huggingface.co · 2025-04-30
AA Intelligence IndexEstimated
~#70 of 103 (estimated)~15.8
- #1Gemini 4 Carbon59.4
- #2Claude Opus 5.557.6
- #69Ring-2.6-1T16.6
- ~#70Phi-4-reasoning-plusEstimated15.8
- #70K2 Horizon 3.7B15.6
No published score — estimated from related benchmarks