Phi-3.5: pricing, context window and benchmarks
Phi-3.5 is a open-weight model from Microsoft. It offers a 16,384-token context window, costs $2.81 per million input tokens and $14.99 per million output tokens, and averages 70.9 across the index's benchmark suite.
Pricing
Phi-3.5 costs $2.81 per million input tokens and $14.99 per million output tokens. A blended 3:1 input-to-output mix works out to about $5.86 per million tokens. Cached input is listed at $0.38 per million tokens.
| Item | Price | Notes |
|---|---|---|
| Input | $2.81 | per 1M tokens |
| Output | $14.99 | per 1M tokens |
| Cached input | $0.38 | per 1M tokens |
| Blended 3:1 | $5.86 | estimate per 1M tokens |
| 1M in + 1M out | $17.80 | 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 | 76.1 | 304 | 49th | |
| GPQA Diamond | 67.0 | 331 | 45th | |
| SWE-bench Verified | 60.9 | 308 | 49th | |
| HumanEval | 76.2 | 327 | 46th | |
| MATH-500 | 75.4 | 316 | 47th | |
| AIME 2025 | 60.6 | 383 | 36th | |
| LiveCodeBench | 59.8 | 371 | 38th | |
| IFEval | 77.1 | 377 | 37th | |
| TruthfulQA | 74.8 | 336 | 44th | |
| HellaSwag | 82.9 | 323 | 46th | |
| MMMU | 68.8 | 328 | 45th | |
| Arena Elo | 1356.7 | 350 | 42th |
What Phi-3.5 is good for
Phi-3.5 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 $2.81 per million input tokens (hosted list price) and a 16,384-token context window, it is a practical option for teams that want more control than a closed API allows.
Strengths
- Input price of $2.81 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
- Context window of 16,384 tokens is small for long-document work.
- Average benchmark score is below the index median; test it on your own task.
How it compares
| Model | Provider | Context | Input / 1M | Avg score |
|---|---|---|---|---|
| Fusion | OpenRouter | 16384 | $8.41 | 71.1 |
| Command A Mini | Cohere | 8192 | $8.12 | 71.0 |
| ERNIE 4.5 Mini | Baidu | 32768 | $8.62 | 71.0 |
| OLMo 2 | Allen Institute for AI | 65536 | $2.99 | 71.0 |
| Auto Turbo | OpenRouter | 65536 | $7.96 | 70.9 |
| Phi-3.5 | Microsoft | 16384 | $2.81 | 70.9 |
FAQ
How much does Phi-3.5 cost?
Phi-3.5 is listed at $2.81 per million input tokens and $14.99 per million output tokens. Cached input is $0.38 per million tokens. A blended 3:1 input-to-output mix is about $5.86 per million tokens.
What is the context window of Phi-3.5?
Phi-3.5 has a context window of 16,384 tokens and can emit up to 16,384 output tokens per request in this index.
Is Phi-3.5 open weights?
Yes. Phi-3.5 is listed as an open-weight model under a open weights licence.
How does Phi-3.5 score on benchmarks?
Phi-3.5 averages 70.9 across the scored benchmarks in this index, ranking #351 of 600 models. Its strongest relative result is Arena Elo at 1356.7.
What is Phi-3.5 best for?
Phi-3.5 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 $2.81 per million input tokens (hosted list price) and a 16,384-token context window, it is a practical option for teams that want more control than a closed API allows.