Zephyr: pricing, context window and benchmarks
Zephyr is a open-weight model from Hugging Face. It offers a 8,192-token context window, costs $1.61 per million input tokens and $6.84 per million output tokens, and averages 62.8 across the index's benchmark suite.
Pricing
Zephyr costs $1.61 per million input tokens and $6.84 per million output tokens. A blended 3:1 input-to-output mix works out to about $2.92 per million tokens. Cached input is listed at $0.19 per million tokens.
| Item | Price | Notes |
|---|---|---|
| Input | $1.61 | per 1M tokens |
| Output | $6.84 | per 1M tokens |
| Cached input | $0.19 | per 1M tokens |
| Blended 3:1 | $2.92 | estimate per 1M tokens |
| 1M in + 1M out | $8.45 | 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 | 68.8 | 408 | 32th | |
| GPQA Diamond | 60.5 | 400 | 33th | |
| SWE-bench Verified | 50.3 | 419 | 30th | |
| HumanEval | 65.9 | 432 | 28th | |
| MATH-500 | 65.3 | 418 | 30th | |
| AIME 2025 | 55.1 | 415 | 31th | |
| LiveCodeBench | 48.7 | 460 | 23th | |
| IFEval | 70.4 | 455 | 24th | |
| TruthfulQA | 67.8 | 417 | 31th | |
| HellaSwag | 76.3 | 423 | 30th | |
| MMMU | 62.2 | 408 | 32th | |
| Arena Elo | 1303.0 | 414 | 31th |
What Zephyr is good for
Zephyr 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 $1.61 per million input tokens (hosted list price) and a 8,192-token context window, it is a practical option for teams that want more control than a closed API allows.
Strengths
- Input price of $1.61 per million tokens is at or below the index median.
- Open weights make self-hosting, fine-tuning and data-residency review possible.
Trade-offs
- Context window of 8,192 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 |
|---|---|---|---|---|
| SmolLM3 | Hugging Face | 8192 | $1.71 | 63.1 |
| DeepSeek-R1 7B | DeepSeek | 32768 | $1.79 | 63.0 |
| Zephyr | Hugging Face | 8192 | $1.61 | 62.8 |
| Hunyuan-Turbo Nano | Tencent | 32768 | $4.65 | 62.7 |
| Kimi K2 7B | Moonshot AI | 16384 | $1.72 | 62.7 |
| GPT-NeoX Instruct | EleutherAI | 16384 | $1.52 | 62.6 |
FAQ
How much does Zephyr cost?
Zephyr is listed at $1.61 per million input tokens and $6.84 per million output tokens. Cached input is $0.19 per million tokens. A blended 3:1 input-to-output mix is about $2.92 per million tokens.
What is the context window of Zephyr?
Zephyr has a context window of 8,192 tokens and can emit up to 8,192 output tokens per request in this index.
Is Zephyr open weights?
Yes. Zephyr is listed as an open-weight model under a open weights licence.
How does Zephyr score on benchmarks?
Zephyr averages 62.8 across the scored benchmarks in this index, ranking #421 of 600 models. Its strongest relative result is Arena Elo at 1303.0.
What is Zephyr best for?
Zephyr 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 $1.61 per million input tokens (hosted list price) and a 8,192-token context window, it is a practical option for teams that want more control than a closed API allows.