Phi-4 Instruct: pricing, context window and benchmarks
Phi-4 Instruct is a multimodal model from Microsoft. It offers a 128,000-token context window, costs $3.79 per million input tokens and $17.11 per million output tokens, and averages 75.3 across the index's benchmark suite.
Pricing
Phi-4 Instruct costs $3.79 per million input tokens and $17.11 per million output tokens. A blended 3:1 input-to-output mix works out to about $7.12 per million tokens. Cached input is listed at $0.34 per million tokens.
| Item | Price | Notes |
|---|---|---|
| Input | $3.79 | per 1M tokens |
| Output | $17.11 | per 1M tokens |
| Cached input | $0.34 | per 1M tokens |
| Blended 3:1 | $7.12 | estimate per 1M tokens |
| 1M in + 1M out | $20.90 | estimate |
Benchmarks
Scores below are modelled estimates on a 0–100 scale. Use them to compare relative position, not as official vendor results.
| Benchmark | Score | Rank | Percentile | Relative |
|---|---|---|---|---|
| MMLU-Pro | 76.6 | 289 | 52th | |
| GPQA Diamond | 73.5 | 186 | 69th | |
| SWE-bench Verified | 62.7 | 263 | 56th | |
| HumanEval | 77.8 | 292 | 51th | |
| MATH-500 | 80.0 | 207 | 66th | |
| AIME 2025 | 74.0 | 182 | 70th | |
| LiveCodeBench | 69.8 | 195 | 68th | |
| IFEval | 80.9 | 314 | 48th | |
| TruthfulQA | 78.2 | 251 | 58th | |
| HellaSwag | 83.4 | 310 | 48th | |
| MMMU | 71.6 | 253 | 58th | |
| Arena Elo | 1401.7 | 205 | 66th |
What Phi-4 Instruct is good for
Phi-4 Instruct is an open-weight model, so it can be self-hosted or run through a third-party inference provider. That matters for data residency, fine-tuning and long-run cost control. At $3.79 per million input tokens (hosted list price) and a 128,000-token context window, it is a practical option for teams that want more control than a closed API allows.
Strengths
- Large 128,000-token context window for long documents and agent traces.
- Input price of $3.79 per million tokens is at or below the index median.
- Open weights make self-hosting, fine-tuning and data-residency review possible.
- Accepts more than text, including image input where the provider supports it.
Trade-offs
- No standout weakness in the indexed specifications, but real-world behaviour still needs testing.
How it compares
| Model | Provider | Context | Input / 1M | Avg score |
|---|---|---|---|---|
| Llama 3.1 Chat | Meta AI | 32768 | $3.82 | 75.3 |
| Llama 4 Scout Base | Meta AI | 32768 | $3.95 | 75.3 |
| Mistral Small 3 | Mistral AI | 32768 | $3.90 | 75.3 |
| Phi-4 Instruct | Microsoft | 128000 | $3.79 | 75.3 |
| Seed-OSS | ByteDance | 32768 | $3.74 | 75.3 |
| Llama 3 Modal Instruct | Modal | 128000 | $3.84 | 75.2 |
FAQ
How much does Phi-4 Instruct cost?
Phi-4 Instruct is listed at $3.79 per million input tokens and $17.11 per million output tokens. Cached input is $0.34 per million tokens. A blended 3:1 input-to-output mix is about $7.12 per million tokens.
What is the context window of Phi-4 Instruct?
Phi-4 Instruct has a context window of 128,000 tokens and can emit up to 4,096 output tokens per request in this index.
Is Phi-4 Instruct open weights?
Yes. Phi-4 Instruct is listed as an open-weight model under a open weights licence.
How does Phi-4 Instruct score on benchmarks?
Phi-4 Instruct averages 75.3 across the scored benchmarks in this index, ranking #244 of 600 models. Its strongest relative result is Arena Elo at 1401.7.
What is Phi-4 Instruct best for?
Phi-4 Instruct is an open-weight model, so it can be self-hosted or run through a third-party inference provider. That matters for data residency, fine-tuning and long-run cost control. At $3.79 per million input tokens (hosted list price) and a 128,000-token context window, it is a practical option for teams that want more control than a closed API allows.