The short answer
Windows in the priced index run from 8,192 tokens to 1,310,720 tokens, and 47 of those 100 priced rows carry a window of 1,000,000 tokens or more. The window is the maximum a model can consider in one request, counted in tokens, but it is not the number that decides your result in Whizi.
The number that decides it is the per turn budget. Every model in the catalogue gets the same flat allowance for one turn: 40,000 input tokens and 20,000 output tokens. A larger window does not raise that budget. A window smaller than 93,000 tokens lowers it.
| Figure | Value |
|---|---|
| Largest window in the priced index | DeepSeek V4 Flash 0731, 1,310,720 tokens |
| Smallest windows in the priced index | The two Tencent Hy-MT2 rows, 8,192 tokens |
| Rows at 1,000,000 tokens or more | 47 of the 100 priced rows |
| Input sent per turn, every catalogue model | 40,000 tokens |
| Output allowed per turn, every catalogue model | 20,000 tokens |
| Window below which that budget shrinks | 93,000 tokens |
So for a real document, the two caps that actually bite are the 100,000 character limit on a single chat message and that 40,000 token input budget. Neither of them is the window column.
Context windows by model, with cost and plan
Every row below comes from the Whizi Cost Index, which prices 100 OpenRouter rows across 29 providers on a standard answer of 1,000 input tokens plus 500 output tokens. The provider rates behind these figures were fetched on 2026-08-20. On the web, the credits column is what one message on that model costs against your credit allowance, and the plan column is the lowest plan that can open it. A platform that is not on the credit allowlist sees a flat 1x instead, is charged one per turn, and keeps the message allowance.
OpenAI.
| Model | Context window | Cost per answer | Credits | Lowest plan |
|---|---|---|---|---|
| GPT-5.6 Luna | 1M | $0.0008 | 1 | Starter |
| GPT-5.4 Nano | 400K | $0.000825 | 2 | Powerhouse |
| GPT-5.4 Mini | 400K | $0.003 | 4 | Powerhouse |
| GPT-5.6 Terra | 1M | $0.008 | 4 | Pro |
| GPT-5.4 | 1M | $0.01 | 15 | Powerhouse |
| GPT-5.6 Sol | 1M | $0.01 | 20 | Powerhouse |
| GPT-5.5 | 1M | $0.02 | 20 | Pro |
| GPT Chat Latest | 400K | $0.02 | 25 | Powerhouse |
Anthropic.
| Model | Context window | Cost per answer | Credits | Lowest plan |
|---|---|---|---|---|
| Claude Haiku 4.5 | 200K | $0.0035 | 4 | Pro |
| Claude Sonnet 5 | 1M | $0.007 | 10 | Pro |
| Claude Sonnet 4.6 | 1M | $0.0105 | 10 | Pro |
| Claude Sonnet 4.5 | 1M | $0.0105 | 10 | Pro |
| Claude Opus 5 | 1M | $0.0175 | 20 | Powerhouse |
| Claude Opus 4.8 | 1M | $0.0175 | 20 | Powerhouse |
| Claude Fable 5 | 1M | $0.035 | 50 | Powerhouse |
Google.
| Model | Context window | Cost per answer | Credits | Lowest plan |
|---|---|---|---|---|
| Gemini 2.5 Flash Lite | 1M | $0.0003 | 1 | Pro |
| Gemini 3.1 Flash Lite | 1M | $0.001 | 1 | Powerhouse |
| Gemini 3.7 Flash | 1M | $0.001313 | 2 | Pro |
| Gemini 2.5 Flash | 1M | $0.00155 | 2 | Pro |
| Gemini 3.5 Flash Lite | 1M | $0.00155 | 2 | Powerhouse |
| Gemini 3.6 Flash | 1M | $0.002625 | 3 | Pro |
| Gemini 3.5 Flash | 1M | $0.006 | 8 | Pro |
| Gemini 3.1 Pro Preview | 1M | $0.008 | 10 | Pro |
DeepSeek.
| Model | Context window | Cost per answer | Credits | Lowest plan |
|---|---|---|---|---|
| DeepSeek V4 Flash 0731 | 1.31M | $0.00028 | 1 | Powerhouse |
| DeepSeek V3.2 | 164K | $0.000469 | 1 | Pro |
| DeepSeek V3.1 | 164K | $0.000725 | 1 | Pro |
| DeepSeek V4 Pro 0813 | 1M | $0.00297 | 2 | Powerhouse |
Qwen.
| Model | Context window | Cost per answer | Credits | Lowest plan |
|---|---|---|---|---|
| Qwen3.7 Flash | 1M | $0.000095 | 1 | Powerhouse |
| Qwen3 Coder Next | 262K | $0.00052 | 1 | Powerhouse |
| Qwen3.6 Flash | 1M | $0.00075 | 1 | Powerhouse |
| Qwen3.7 Plus | 1M | $0.00096 | 1 | Pro |
| Qwen3.8 27B | 1M | $0.00205 | 3 | Powerhouse |
| Qwen3.7 Max | 1M | $0.003688 | 5 | Powerhouse |
| Qwen3.8 Max | 1M | $0.005 | 6 | Powerhouse |
xAI.
| Model | Context window | Cost per answer | Credits | Lowest plan |
|---|---|---|---|---|
| Grok Build 0.1 | 256K | $0.002 | 3 | Powerhouse |
| Grok 4.3 | 1M | $0.0025 | 4 | Powerhouse |
| Grok 4.5 | 500K | $0.005 | 6 | Powerhouse |
| Grok 4.6 | 500K | $0.005 | 6 | Pro |
Meta.
| Model | Context window | Cost per answer | Credits | Lowest plan |
|---|---|---|---|---|
| Llama 3.3 70B Instruct | 131K | $0.00026 | 1 | Pro |
| Llama 4 Maverick | 1M | $0.0006 | 1 | Pro |
| Muse Glimmer 30B | 131K | $0.0011 | 2 | Powerhouse |
| Muse Spark 1.2 | 1M | $0.003375 | 5 | Powerhouse |
Mistral.
| Model | Context window | Cost per answer | Credits | Lowest plan |
|---|---|---|---|---|
| Codestral 2508 | 256K | $0.00075 | 1 | Powerhouse |
| Mistral Large 3 2512 | 262K | $0.00125 | 2 | Pro |
| Mistral Medium 3.1 | 131K | $0.0014 | 2 | Pro |
| Mistral Medium 3.5 | 262K | $0.00525 | 6 | Powerhouse |
Z.ai and Moonshot.
| Model | Context window | Cost per answer | Credits | Lowest plan |
|---|---|---|---|---|
| GLM 4.6 | 205K | $0.0015 | 1 | Powerhouse |
| GLM 4.7 | 205K | $0.001275 | 2 | Powerhouse |
| GLM 5 | 205K | $0.00156 | 2 | Pro |
| GLM 5.2 | 1M | $0.002484 | 3 | Powerhouse |
| GLM 5.3 | 1M | $0.0036 | 5 | Powerhouse |
| Kimi K2 0711 | 131K | $0.00172 | 10 | Powerhouse |
| Kimi K2 Thinking | 262K | $0.00185 | 10 | Powerhouse |
| Kimi K2.7 Code | 262K | $0.00246 | 10 | Powerhouse |
| Kimi K3 | 1M | $0.0105 | 10 | Pro |
A further fourteen rows.
These are a selection of the priced rows from the remaining labs, not all of them.
| Model | Context window | Cost per answer | Credits | Lowest plan |
|---|---|---|---|---|
| MiniMax M2 | 205K | $0.000765 | 1 | Powerhouse |
| MiniMax M3 | 1M | $0.0009 | 1 | Powerhouse |
| Nemotron 3.5 Lightning, NVIDIA | 262K | $0.00018 | 1 | Powerhouse |
| Nemotron 3 Ultra, NVIDIA | 512K | $0.0024 | 3 | Powerhouse |
| Nova Pro 1.0, Amazon | 300K | $0.0024 | 3 | Powerhouse |
| Command A, Cohere | 256K | $0.0075 | 10 | Powerhouse |
| Solar Pro 4, Upstage | 524K | $0.00009 | 1 | Powerhouse |
| Ling-3.0-flash, inclusionAI | 262K | $0.000053 | 1 | Powerhouse |
| Nex-N2-Mini, Nex Agi | 262K | $0.000075 | 1 | Powerhouse |
| Laguna XS 2.1, Poolside | 262K | $0.00012 | 1 | Powerhouse |
| Granite 4.1 8B, IBM | 131K | $0.0001 | 1 | Powerhouse |
| Phi 4, Microsoft | 16K | $0.00014 | 1 | Powerhouse |
| Inkling, Thinkingmachines | 1M | $0.002975 | 4 | Powerhouse |
| Fugu Ultra, Sakana | 1M | $0.02 | 25 | Powerhouse |
The catalogue holds 280+ models, and the rows above are drawn from the Cost Index, which prices 100 OpenRouter rows for comparison, and neither the tables nor the index are the whole catalogue. The plan column is derived rather than listed, and the model list covers how that derivation works.
Starter is the one plan the table barely covers. Its four entries are Auto, the house model Whizi AI, the base GPT model that Whizi AI runs on, and one fast Gemini model. Whizi AI is a house persona on that base GPT model rather than a model of its own, and it costs 1 credit a message. Free accounts are not model gated at all: a free account can try any text model inside a lifetime allowance of 7 messages.
The windows small enough to change the budget
The window column is not the deciding column for length, because the per turn budget is flat across the catalogue. How that budget is built, and what happens when a conversation runs past it, is covered in when the conversation is too long.
What a window does decide is whether that flat budget applies at all, and the line sits at 93,000 tokens. Four index rows named on this page fall under it, and two of them are the smallest windows in the whole priced index.
| Row | Context window | Printed in the tables above |
|---|---|---|
| Tencent Hy-MT2, two rows | 8,192 | No |
| Phi 4, Microsoft | 16K | Yes |
| DeepSeek R1 | 64K | No |
What a window means for a real document
The first cap a document hits is not the context window. It is the 100,000 character limit on a single message, and text extracted from an attached file counts against it exactly like typed text.
On the web, PDF, Word and spreadsheet files are extracted to text in your browser before anything is sent, so what the model receives is that extracted text. Each file is capped at 10 MB. The full format list is in supported file types.
Projects are the one place a document really does eat the budget, because a pinned project file is rebuilt into the prompt on every single turn in that project, inside the same flat input budget no matter which window the model carries. The pinned file and project block caps are in when the conversation is too long.
Window against price and credits, in the rows above
Sort the tables by window and nothing else sorts with them. The largest window in the priced index, at 1,310,720 tokens, costs $0.00028 per standard answer, charges 1 credit a message and sits on Powerhouse. The cheapest row of all costs $0.000053 per answer on a 262K window. The median row sits at $0.00185 per answer, also on a 262K window. The dearest row costs $0.105 per standard answer and carries no credit rung at all, because it is not one of the rows Whizi offers.
The scatter is easiest to read on rows that share a window.
| Row | Context window | Cost per answer | Credits |
|---|---|---|---|
| DeepSeek V4 Flash 0731 | 1.31M | $0.00028 | 1 |
| Qwen3.7 Flash | 1M | $0.000095 | 1 |
| Inkling, Thinkingmachines | 1M | $0.002975 | 4 |
| Claude Fable 5 | 1M | $0.035 | 50 |
| Granite 4.1 8B, IBM | 131K | $0.0001 | 1 |
| Kimi K2 0711 | 131K | $0.00172 | 10 |
Three rows there carry the same 1,000,000 token window and charge 1, 4 and 50 credits. Two rows carry the same 131K window and charge 1 and 10. The priced index spans roughly 2000x from cheapest row to dearest, and the rows above show that spread does not follow the window column.
- Windows in the priced index run from 8,192 tokens to 1,310,720 tokens
- 47 of the 100 priced rows carry a window of 1,000,000 tokens or more
- The largest window in the index answers once for $0.00028 and charges 1 credit
- The cheapest row in the index carries a 262K window, not the largest one
- The median priced row sits at $0.00185 per standard answer, also on a 262K window
- On the web, the credits column is what one message spends against the credit allowance
- A credit rung is derived from provider per token rates, so window size is not an input to it
- Whizi AI is a house persona on a base GPT model, at 1 credit a message
- Free accounts are not model gated: any text model, inside 7 lifetime messages
- The Cost Index prices 100 OpenRouter rows, which is not the whole 280+ catalogue
Frequently asked questions
Do I get a bigger context window on a higher plan?
No. The window belongs to the model, not to your subscription. What a higher plan buys is access to more models and a larger monthly credit allowance: 400 credits on Starter, 2,000 on Pro and 8,000 on Powerhouse.
Does a model with a bigger context window cost more credits?
Not in the tables above. The largest window in the priced index, at 1,310,720 tokens, charges 1 credit a message, while a 131K window row in the same tables charges 10. A credit rung is derived from a reference turn of 3,000 input tokens and 800 output tokens against live provider per token rates, then rounded up to a legal rung, and window size is not one of the inputs.
Can I try a large context window model on a free account?
Yes. The free tier is deliberately not model gated, so a free account can open any text model in the catalogue, including the 1,000,000 token rows. The limit is the allowance rather than the model list: 7 messages, lifetime, which does not reset.
Does Whizi publish a context window for every model in the catalogue?
No. The windows on this page come from the Cost Index, which prices 100 OpenRouter rows for comparison, and the catalogue holds 280+ models. The picker in the app is a separate surface served live by the shared worker, so it lists the current catalogue rather than this index.
Where do the context window figures on this page come from?
From the Whizi Cost Index, which prices 100 OpenRouter rows on a standard answer of 1,000 input tokens plus 500 output tokens, with source prices fetched from OpenRouter on 2026-08-20. The Cost Index records that single fetch date, so treat the dollar figures as of that day rather than as live rates.