Phi-4 14B
consumerAI-generated content marks
The provider reports that this model does not add embedded watermarks or provenance metadata to generated output.
- 14.7B
- 16k
- mit
- 🇺🇸 USA
- Dec 2024
Scores
Score per dollar
1000 pts per $/M input
general_score (70) divided by cheapest input price ($0.07/M). Higher is better value. See live pricing.
Related models
Save your hardware and every model page answers the real question: will it run on your machine, and how fast?
Join free - save your rig →Run it locally
Per-quant memory needs and a static "can you run it?" reference - no rig entry required
The reference hardware
22 reference configs, drawn in-house. Scroll for more.
Can you run it? - reference rigs
| Rig | Q4_K_M | Q8_0 | FP16 |
|---|---|---|---|
| 4x H100 80GB (320GB) | fast 726.8t/s | fast 442.9t/s | fast 240.5t/s |
| NVIDIA DGX Station 748GB | fast 433.9t/s | fast 264.4t/s | fast 143.6t/s |
| 8x RTX 3090 rack (192GB) | fast 406.2t/s | fast 247.6t/s | fast 134.4t/s |
| 4x RTX 5090 (128GB) | fast 388.8t/s | fast 236.9t/s | fast 128.7t/s |
| AMD Instinct MI300X (192GB) | fast 288.8t/s | fast 176.0t/s | fast 95.6t/s |
| 4x RTX 4090 (96GB) | fast 218.7t/s | fast 133.3t/s | fast 72.4t/s |
| 2x RTX 5090 (64GB) | fast 194.4t/s | fast 118.5t/s | fast 64.3t/s |
| 2x RTX 3090 (48GB) | fast 101.6t/s | fast 61.9t/s | fast 33.6t/s |
| Single RTX 5090 (32GB) | fast 97.2t/s | fast 59.2t/s | fast 32.2t/s |
| RTX PRO 6000 Blackwell (96GB) | fast 97.2t/s | fast 59.2t/s | fast 32.2t/s |
| Mac Studio M4 Ultra 192GB | fast 64.6t/s | fast 39.4t/s | fast 21.4t/s |
| Mac Studio M4 Ultra 512GB | fast 64.6t/s | fast 39.4t/s | fast 21.4t/s |
| Single RTX 4090 (24GB) | fast 54.7t/s | fast 33.3t/s | offload |
| MacBook Pro M5 Max 128GB | fast 36.3t/s | fast 22.1t/s | ok 12.0t/s |
| Single GTX 1080 Ti (11GB) | fast 26.3t/s | offload | no -> cloud |
| Dual EPYC 9004 + 768GB DDR5-4800 | fast 25.0t/s | ok 15.2t/s | ok 8.3t/s |
| DGX Spark 128GB unified | ok 14.8t/s | ok 9.0t/s | slow 4.9t/s |
| Ryzen AI Max+ 395 128GB | ok 13.9t/s | ok 8.5t/s | slow 4.6t/s |
| Jetson AGX Orin 64GB | ok 11.1t/s | slow 6.8t/s | slow 3.7t/s |
| Epyc + 512GB DDR4-3200 + 2x RTX 3090 | ok 11.1t/s | slow 6.8t/s | slow 3.7t/s |
| Epyc + 512GB DDR4-2400 + 2x RTX 3090 | ok 8.3t/s | slow 5.1t/s | slow 2.8t/s |
| NVIDIA Jetson Orin NX 16GB | slow 5.6t/s | tight | no -> cloud |
Fit tiers use the same will-it-run logic as the rig finder. For comfortable fits, the badge reflects decode speed: fast >=20 t/s, ok 8-20 t/s, slow <8 t/s. t/s is a bandwidth estimate, not a measured benchmark.
Download options
Or run it in the cloud
Live per-provider pricing, throughput and uptime - refreshed about 13 hours ago via OpenRouter. Click a column to sort.
| Provider | Type | Input $/M | Output $/M | Cache $/M | Tok/s | Latency | Uptime | Value |
|---|---|---|---|---|---|---|---|---|
|
DeepInfra
|
API | 0.07 | 0.14 | - | - | - | 100.00% | best uptime |
Default order: throughput among 95%+ uptime providers, then latency; subscriptions last. Sort by any column. Subscription rows show $/mo in the Value column - per-token columns are "-". Affiliate links are marked sponsored / nofollow. Confirm current pricing on the provider's site before committing.
Detailed API pricing page + JSON endpoint →
See who runs Microsoft in production →
Inference cost over time
Data accumulates from the first daily sync - longer ranges populate over time. Prices come from OpenRouter snapshots, not a historical API.