Phi-4 8B: pricing, context window and benchmarks
Phi-4 8B is a small open-weight model from Microsoft. It offers a 32,768-token context window, costs $0.96 per million input tokens and $5.09 per million output tokens, and averages 54.9 across the index's benchmark suite.
Pricing
Phi-4 8B costs $0.96 per million input tokens and $5.09 per million output tokens. A blended 3:1 input-to-output mix works out to about $1.99 per million tokens. Cached input is listed at $0.12 per million tokens.
| Item | Price | Notes |
|---|---|---|
| Input | $0.96 | per 1M tokens |
| Output | $5.09 | per 1M tokens |
| Cached input | $0.12 | per 1M tokens |
| Blended 3:1 | $1.99 | estimate per 1M tokens |
| 1M in + 1M out | $6.05 | 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 | 61.9 | 504 | 16th | |
| GPQA Diamond | 53.2 | 489 | 19th | |
| SWE-bench Verified | 39.9 | 523 | 13th | |
| HumanEval | 57.1 | 531 | 12th | |
| MATH-500 | 54.8 | 522 | 13th | |
| AIME 2025 | 43.5 | 518 | 14th | |
| LiveCodeBench | 41.8 | 527 | 12th | |
| IFEval | 65.8 | 523 | 13th | |
| TruthfulQA | 62.2 | 501 | 17th | |
| HellaSwag | 70.9 | 523 | 13th | |
| MMMU | 52.5 | 524 | 13th | |
| Arena Elo | 1256.7 | 480 | 20th |
What Phi-4 8B is good for
Phi-4 8B is a small model. That makes it a good fit for classification, extraction, routing, summarization and other high-volume jobs where latency and cost dominate. It is not the right pick for open-ended frontier reasoning, but at $0.96 per million input tokens it can handle a large share of production traffic.
Strengths
- Input price of $0.96 per million tokens is at or below the index median.
- Open weights make self-hosting, fine-tuning and data-residency review possible.
Trade-offs
- Average benchmark score is below the index median; test it on your own task.
- Less reliable on open-ended reasoning and complex multi-step instructions.
How it compares
| Model | Provider | Context | Input / 1M | Avg score |
|---|---|---|---|---|
| Llama 4 Maverick 1.5B | Meta AI | 8192 | $0.93 | 55.2 |
| Yi-Large 7B | 01.AI | 4096 | $0.83 | 55.0 |
| Phi-4 8B | Microsoft | 32768 | $0.96 | 54.9 |
| Mixtral Groq 8B | Groq | 32768 | $0.77 | 54.8 |
| GLM-4.6 1.5B | Zhipu AI | 4096 | $0.92 | 54.7 |
| MiniMax-M2 1.5B | MiniMax | 16384 | $0.84 | 54.6 |
FAQ
How much does Phi-4 8B cost?
Phi-4 8B is listed at $0.96 per million input tokens and $5.09 per million output tokens. Cached input is $0.12 per million tokens. A blended 3:1 input-to-output mix is about $1.99 per million tokens.
What is the context window of Phi-4 8B?
Phi-4 8B has a context window of 32,768 tokens and can emit up to 32,768 output tokens per request in this index.
Is Phi-4 8B open weights?
Yes. Phi-4 8B is listed as an open-weight model under a open weights licence.
How does Phi-4 8B score on benchmarks?
Phi-4 8B averages 54.9 across the scored benchmarks in this index, ranking #524 of 600 models. Its strongest relative result is Arena Elo at 1256.7.
What is Phi-4 8B best for?
Phi-4 8B is a small model. That makes it a good fit for classification, extraction, routing, summarization and other high-volume jobs where latency and cost dominate. It is not the right pick for open-ended frontier reasoning, but at $0.96 per million input tokens it can handle a large share of production traffic.