phi-4
7 workload results across 3 hardware configurations.
Fastest community-submitted config (not measured by llm-speed)
140.9 decode tok/s
on RTX 5090 (32GB) via llama.cpp. see full run
For a standard 4-bit quantization using MLX, here are measured results for short-chat generation.
On MLX 4-bit, phi-4 is fastest on the M3 Ultra (60-core GPU) at 74.37 tok/s, 376% faster than the M3 Pro (18-core GPU) (15.61 tok/s).
| Card | Decode tok/s | TTFT (113-token prompt) | Run | Source |
|---|---|---|---|---|
| M3 Ultra (60-core GPU) | 74.37 (single run) | 443 ms | r_sqzp-0rdez- | community-submitted |
| M3 Pro (18-core GPU) | 15.61 (single run) | 936 ms | r_-w8hnn61va_ | community-submitted |
phi-4, 4-bit, MLX, chat-short, batch 1. Only cards with their standard memory are listed; CPU-only hosts are excluded. Community-submitted runs are signed on the submitter's own machine and are not measured by llm-speed; llm-speed verdicts use only its own runs.
Signed runs (7 workload results)
All 4 runs are community-submitted (signed on the submitter's own machine, not measured by llm-speed, and not used for llm-speed verdicts).
RTX 5090 (32GB)
| Workload | Backend | Quant | decode tok/s | prefill tok/s | TTFT | Run | Source |
|---|---|---|---|---|---|---|---|
| chat-short | llama.cpp | - | 139.1tok/s | no data | 51.1ms | r_k1u_k1j_1i2 | Community-submitted |
| chat-long | llama.cpp | - | 133.7tok/s | no data | 433ms | r_k1u_k1j_1i2 | Community-submitted |
| concurrent-decode | llama.cpp | - | 140.2tok/s | no data | no data | r_k1u_k1j_1i2 | Community-submitted |
| agent-trace | llama.cpp | - | 124.8tok/s | 21,162.4tok/s | 102ms | r_k1u_k1j_1i2 | Community-submitted |
| chat-short | llama.cpp | - | 140.9tok/s | no data | 56.7ms | r_e-k4aea8ipr | Community-submitted |
M3 Pro (18-core GPU) + 36GB unified
| Workload | Backend | Quant | decode tok/s | prefill tok/s | TTFT | Run | Source |
|---|---|---|---|---|---|---|---|
| chat-short | mlx@0.31.3 | 4BIT | 15.61tok/s | 120.7tok/s | 936ms | r_-w8hnn61va_ | Community-submitted |
M3 Ultra (60-core GPU) + 96GB unified
| Workload | Backend | Quant | decode tok/s | prefill tok/s | TTFT | Run | Source |
|---|---|---|---|---|---|---|---|
| chat-short | mlx@0.31.3 | 4BIT | 74.37tok/s | 255.0tok/s | 443ms | r_sqzp-0rdez- | Community-submitted |
Can phi-4 run on your hardware?
Fastest decode speed on each GPU or Mac, from llm-speed's own runs and community-submitted runs. Each page lists every quant and runtime and labels which runs llm-speed measured.
phi-4 on hardware
Request a phi-4 run
No measured run yet for these pairs. Vote for the one you want; the most-requested runs go first. See all requests.
We never sell or share your email. Privacy
Compare phi-4 head-to-head
Check it for your setup
- Predict your exact setup: estimate whether phi-4 fits your GPU or Mac at your quant and context, next to measured speeds where we have them.
- Request this pair: ask for a measured phi-4 run on hardware we have not measured. Requests are ranked by votes.
- Full run table (4 runs): every phi-4 result we have, with links to each source run.