> ## 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 Request Ratios > What each model costs against your plan allowance, and how many requests that buys on every plan. # Model Request Ratios Not every model costs the same slice of your plan. Each model has a **ratio** — how much of your allowance one call to it uses. **Auto and flash models are the 1.0 floor.** GLM-5.2 costs **2.0** — one call uses two requests' worth. Your plan's "requests per window" number is what that buys on Auto and other 1.0 models; GLM-5.2 calls consume twice as much. On **Free** (**25** per 5-hour window) that is **25** Auto calls or **12** GLM-5.2 calls; a very long GLM-5.2 prompt costs **4.0** (down to **6** on Free). > **Note:** mid-tier models sit between 1.0 and 2.0; premium chat models cost > more — see the table below. GLM-5.2 long-context surcharges are > separate — see [Long-context surcharge](#long-context-surcharge). Image models > use a fixed ratio per image — see [Image generation ratios](#image-generation-ratios). > **Promo ratio:** `Qwen3.8 27b` is **0.5** through **2026-09-30** UTC (standard ratio ~~1~~). > **Promo ratio:** `DeepSeek-V4-Flash-0731` is **0.75** through **2026-09-30** UTC (standard ratio ~~1~~). ## Ratios at a glance | Ratio | Models | | -------------- | ----------------------------------------------------------- | | **4** | `GLM-5.2`, `GLM-5.3` when the prompt exceeds 262,144 tokens | | **2** | `GLM-5.2`, `GLM-5.3` | | **1.7** | `DeepSeek-V4-Pro-0813` | | **1.6** | `Kimi K2.7 Code` | | **1** | `Auto`, `GLM-5.3-Flash`, `MiniMax M3` | | ~~1~~ **0.75** | `DeepSeek-V4-Flash-0731` | Use the model name exactly as shown as the `model` value in your request. Tables below list models on subscription plans (Free through Max+). On-demand-only models (for example Kimi K3) are billed per token and are not shown here. > **Legacy Auto Agent plans:** Existing subscribers on Auto Agent Mini, Lite, > Pro, or Max keep their grandfathered allowance at the same ratios until they > choose to switch plans. ## How many requests you get ### Per 5-hour window | Model | Ratio | Free | Lite | Pro | Pro+ | Max | Max+ | | ------------------------------ | -------------- | ------------- | --------------- | ----------------- | ------------------- | ------------------- | ------------------- | | **GLM-5.2** (prompt > 262,144) | **4** | 6 | 100 | 200 | 300 | 400 | 800 | | **GLM-5.3** (prompt > 262,144) | **4** | 6 | 100 | 200 | 300 | 400 | 800 | | GLM-5.2 | 2 | 12 | 200 | 400 | 600 | 800 | 1,600 | | GLM-5.3 | 2 | 12 | 200 | 400 | 600 | 800 | 1,600 | | DeepSeek-V4-Pro-0813 | 1.7 | 14 | 235 | 470 | 705 | 941 | 1,882 | | Kimi K2.7 Code | 1.6 | 15 | 250 | 500 | 750 | 1,000 | 2,000 | | Auto | 1 | 25 | 400 | 800 | 1,200 | 1,600 | 3,200 | | GLM-5.3-Flash | 1 | 25 | 400 | 800 | 1,200 | 1,600 | 3,200 | | MiniMax M3 | 1 | 25 | 400 | 800 | 1,200 | 1,600 | 3,200 | | DeepSeek-V4-Flash-0731 | ~~1~~ **0.75** | ~~25~~ **33** | ~~400~~ **533** | ~~800~~ **1,066** | ~~1,200~~ **1,600** | ~~1,600~~ **2,133** | ~~3,200~~ **4,266** | ### Per week Weekly caps reset Monday 00:00 UTC. | Model | Ratio | Free | Lite | Pro | Pro+ | Max | Max+ | | ------------------------------ | -------------- | --------------- | -------------------- | --------------------- | --------------------- | --------------------- | --------------------- | | **GLM-5.2** (prompt > 262,144) | **4** | 87 | 1,875 | 3,250 | 4,875 | 6,500 | 13,000 | | **GLM-5.3** (prompt > 262,144) | **4** | 87 | 1,875 | 3,250 | 4,875 | 6,500 | 13,000 | | GLM-5.2 | 2 | 175 | 3,750 | 6,500 | 9,750 | 13,000 | 26,000 | | GLM-5.3 | 2 | 175 | 3,750 | 6,500 | 9,750 | 13,000 | 26,000 | | DeepSeek-V4-Pro-0813 | 1.7 | 205 | 4,411 | 7,647 | 11,470 | 15,294 | 30,588 | | Kimi K2.7 Code | 1.6 | 218 | 4,687 | 8,125 | 12,187 | 16,250 | 32,500 | | Auto | 1 | 350 | 7,500 | 13,000 | 19,500 | 26,000 | 52,000 | | GLM-5.3-Flash | 1 | 350 | 7,500 | 13,000 | 19,500 | 26,000 | 52,000 | | MiniMax M3 | 1 | 350 | 7,500 | 13,000 | 19,500 | 26,000 | 52,000 | | DeepSeek-V4-Flash-0731 | ~~1~~ **0.75** | ~~350~~ **466** | ~~7,500~~ **10,000** | ~~13,000~~ **17,333** | ~~19,500~~ **26,000** | ~~26,000~~ **34,666** | ~~52,000~~ **69,333** | ## Image generation ratios Each successful image counts as one call. Ratios apply per image, not per token. | Model | Ratio | | -------------- | ----- | | SDXL Base 1.0 | 25 | | SDXL Lightning | 25 | You do not have to spend a window on one model — costs simply add up. On Pro, 600 GLM-5.2 calls (1,200) plus 400 Auto calls (400) is 1,600 requests' worth. ## Long-context surcharge **A GLM-5.2 request whose prompt exceeds 262,144 tokens costs 4.0 instead of 2.0.** * GLM-5.2 is the **only** chat model with a context surcharge. * It is measured on **input** only. A long *reply* to a short prompt still costs 2.0. * The threshold is measured on **upstream-reported input tokens** after the call completes. `X-Quota-Cost` on the response uses a pre-flight estimate and may read lower until settlement; the dashboard is always exact. If you routinely work above 262k tokens, `DeepSeek-V4-Pro-0813` (alias `DeepSeek-V4-Pro`) and `MiniMax M3` both have larger context windows and carry **no** surcharge. ## Reading the cost at runtime You never have to hardcode this table. `GET /v1/models` reports the ratio on every model: ```bash theme={null} curl https://api.openference.com/v1/models \ -H "Authorization: Bearer YOUR_API_KEY" ``` ```json theme={null} { "id": "GLM-5.2", "quota_multiplier": 2, "quota_context_surcharge": { "threshold_input_tokens": 262144, "factor": 2 } } ``` `quota_context_surcharge` is present only on models whose cost scales with prompt size. Every response also carries what that specific call used: | Header | Meaning | | ------------------- | ---------------------------------------------------------- | | `X-Quota-Cost` | What this request cost, in requests (e.g. `1`, `2`, `2.5`) | | `X-Quota-Remaining` | Approximately what is left in the current window | > **Note:** `X-Quota-Remaining` is read when the request is admitted, so with > several requests in flight at once it can read slightly high. The dashboard > under **Billing** is always exact. ## What still counts as one request The ratio changes how much a call costs, not which calls count: * `success` and `client_error` (4xx) count. * **Upstream provider errors and capacity-unavailable situations do not count** — a failed request is refunded in full, at whatever it would have cost. * The **per-minute burst limit** is unaffected. It is a hard cap on calls per minute and is never extended by credits. ## Related * [Plans & Usage](/billing/plans) — allowances, weekly caps, burst limits * [Pricing & Transparency](/models/pricing) — on-demand token rates beyond your allowance * [Model Catalog](/models/catalog) — context windows and capabilities