> ## Documentation Index
> Fetch the complete documentation index at: https://docs.openference.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Model Costs & Usage Estimates

> Estimated requests per window on Minion ($20/month) and Boss ($80/month) at 10,000 input + 100 output tokens, plus model token rates for plans and on-demand usage.

# Model Costs & Usage Estimates

## Usage limits

Usage is tracked across three rolling windows:

* **5-hour limit (20%)** — Fixed daily windows (00:00, 05:00, 10:00, 14:00, and 19:00 UTC).
* **Weekly limit (50%)** — Resets every Monday at 14:00 UTC.
* **Monthly limit (100%)** — Resets on your monthly billing date.

### Estimated requests (Minion)

Estimated request counts based on a typical coding turn of **10,000 input and 100 output tokens** (uncached). Active prompt caching meters repeated context at the much lower **Cached Read** rate, enabling significantly more requests.

| Model | Requests per 5 hours | Requests per week | Requests per month |
| - | - | - | - |
| `Kimi K3` | 380 | 950 | 1,900 |
| `GLM-5.3` | 830 | 2,080 | 4,160 |
| `DeepSeek-V4-Pro-0813` | 880 | 2,210 | 4,410 |
| `DeepSeek-V4-Flash-0731` | 2,650 | 6,620 | 13,240 |
| `DeepSeek-V4.1-Flash` | 3,850 | 9,620 | 19,230 |
| `DeepSeek-V4.1-Flash-1017` | 3,850 | 9,620 | 19,230 |
| `GLM-5.3-Flash` | 7,740 | 19,350 | 38,710 |
| `GLM-4.7-Flash` | 18,750 | 46,880 | 93,750 |
| `Agnes-3.0-Flash` | 23,300 | 58,250 | 116,500 |
| `Auto` | — | — | — |
| `Gemma 4 26B` | Unlimited | Unlimited | Unlimited |
| `GPT-OSS-120B` | Unlimited | Unlimited | Unlimited |
| `Llama 3.2 3B` | Unlimited | Unlimited | Unlimited |
| `MiniMax M3` | Unlimited | Unlimited | Unlimited |
| `Nemotron-3-120B` | Unlimited | Unlimited | Unlimited |
| `Qwen3.8 27b` | Unlimited | Unlimited | Unlimited |

### Model token rates (Minion, Boss & On-Demand)

Token prices are per 1M tokens.

| Model | Input (/ 1M tokens) | Output (/ 1M tokens) | Cached Read (/ 1M tokens) |
| - | - | - | - |
| `Kimi K3` | \$3.00 | \$15.00 | \$0.30 |
| `GLM-5.3` | \$1.40 | \$4.40 | \$0.26 |
| `DeepSeek-V4-Pro-0813` | \$1.32 | \$3.96 | \$0.132 |
| `DeepSeek-V4-Flash-0731` | \$0.44 | \$1.32 | \$0.044 |
| `DeepSeek-V4.1-Flash` | \$0.30 | \$1.20 | \$0.006 |
| `DeepSeek-V4.1-Flash-1017` | \$0.30 | \$1.20 | \$0.006 |
| `GLM-5.3-Flash` | \$0.15 | \$0.50 | \$0.03 |
| `GLM-4.7-Flash` | \$0.06 | \$0.40 | — |
| `Agnes-3.0-Flash` | \$0.05 | \$0.15 | \$0.005 |
| `Auto` | \$0.3675 | \$1.00 | — |
| `Gemma 4 26B` | \$0.10 | \$0.30 | — |
| `GPT-OSS-120B` | \$0.35 | \$0.75 | — |
| `Llama 3.2 3B` | \$0.051 | \$0.335 | — |
| `MiniMax M3` | \$0.30 | \$1.20 | \$0.06 |
| `Nemotron-3-120B` | \$0.50 | \$1.50 | — |
| `Qwen3.8 27b` | \$0.45 | \$3.20 | — |

## What counts toward your allowance

* `success` and `client_error` (4xx) requests count toward your plan allowance.
* **Upstream provider errors and capacity-unavailable situations never count** against your allowance.
* The **per-minute burst limit** is a separate hard cap on requests per minute and is not extended by credits.

> **On a retired legacy plan?** Existing subscribers on **Promo**, **Lite**, **Pro**, **Pro+**, **Max**, or **Max+** use per-model request ratios until the end of their current billing period — see [Model Request Ratios (Legacy)](https://docs.openference.com/getting-started/model-ratios.md).

## Related

* [Minion Plan](https://docs.openference.com/billing/minion.md) — \$20/month usage windows and request estimates
* [Boss Plan](https://docs.openference.com/billing/boss.md) — \$80/month usage windows and request estimates
* [Plans & Usage](https://docs.openference.com/billing/plans.md) — how plan windows, resets, and on-demand credits work
* [Pricing & Transparency](https://docs.openference.com/models/pricing.md) — on-demand token rates beyond your allowance
* [Model Request Ratios (Legacy)](https://docs.openference.com/getting-started/model-ratios.md) — legacy request ratios and retired plan limits


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.