Context window costs
A context window is a ceiling, not a target. It tells you what a model will accept. It says nothing about what you should send, and the price of sending it climbs in a straight line with every token you add.
This page prices that line. It takes four input sizes and charges every model in the catalogue for reading each one once.
What it costs to fill a context window
337 models, snapshot 2026-09-15. Regenerated from the models.dev catalogue by apps/web/scripts/fetch-models.ts on every build, not maintained by hand.
| Context | Models | Cheapest | Median | Dearest |
|---|---|---|---|---|
| 320 of 337 | $0.0015 | $0.029 | $14.00 | |
| 190 of 337 | $0.0035 | $0.069 | $7.09 | |
| 100 of 337 | $0.0075 | $0.200 | $15.00 | |
| 96 of 337 | $0.015 | $0.400 | $30.00 |
Pick a size to price it against every model below. 236K is the median repository bundle measured across 60 public repositories, not a round number.
190 models accept 236,218 tokens
| Model | Per 1M in | To fill |
|---|---|---|
| GLM-5.3-FlashMerge Gateway | $0.015 | $0.0035 |
| Qwen FlashAlibaba (China) | $0.022 | $0.0052 |
| Seed 1.6 FlashVolcengine Ark | $0.022 | $0.0053 |
| Qwen3.7 FlashAIHubMix | $0.028 | $0.0067 |
| Qwen3.5 FlashMerge Gateway | $0.029 | $0.0069 |
| Seed 2.0 MiniVolcengine Ark | $0.030 | $0.0070 |
| Solar Pro 4LLMTR | $0.030 | $0.0071 |
| DeepSeek V4 FlashMerge Gateway | $0.035 | $0.0083 |
| DeepSeek V4 Flash 0731Merge Gateway | $0.035 | $0.0083 |
| GPT-5 MiniQiHang | $0.040 | $0.0094 |
| Qwen3.5 9BCrofAI | $0.040 | $0.0094 |
| Gemma 4 26B A4B ITKilo Gateway | $0.042 | $0.0099 |
| Qwen TurboAlibaba (China) | $0.044 | $0.010 |
| GPT-5 NanoJiekou.AI | $0.045 | $0.011 |
| Nemotron 3 Nano 30B A3BCrusoe | $0.050 | $0.012 |
| Nemotron 3.5 Lightning 30B A3BRunInfra | $0.050 | $0.012 |
| Gemini 2.0 Flash-LitePoe | $0.052 | $0.012 |
| Qwen3.5 35B-A3BOrcaRouter | $0.057 | $0.013 |
| GPT-5.6 LunaBothub | $0.060 | $0.014 |
| Laguna XS 2.1OpenRouter | $0.060 | $0.014 |
| Ling 3.0 Flash FinOpenRouter | $0.060 | $0.014 |
| Nova LiteAmazon Bedrock | $0.060 | $0.014 |
| Qwen3-Coder 30B-A3B InstructTensorX | $0.060 | $0.014 |
| Hy3 previewSiliconFlow | $0.066 | $0.016 |
| Gemini 3 Flash PreviewQiHang | $0.070 | $0.017 |
| Qwen3.6 35B-A3BEmpirioLabs AI | $0.070 | $0.017 |
| DeepSeek V4 Flash Vision ExpCrofAI | $0.080 | $0.019 |
| GPT-4.1 nanoSAP AI Core | $0.080 | $0.019 |
| Nemotron 3 Super 120B A12BKilo Gateway | $0.080 | $0.019 |
| Hy3Kilo Gateway | $0.083 | $0.019 |
| Qwen3.5 27BOrcaRouter | $0.086 | $0.020 |
| Qwen3 235B-A22B Instruct 2507OpenRouter | $0.087 | $0.021 |
| Seed 2.0 LiteVolcengine Ark | $0.089 | $0.021 |
| Gemini 2.5 FlashQiHang | $0.090 | $0.021 |
| Laguna S 2.1OpenRouter | $0.090 | $0.021 |
| Devstral Small 2Pioneer | $0.100 | $0.024 |
| Gemini 2.0 FlashPoe | $0.100 | $0.024 |
| Gemini 2.5 Flash-LiteLLMTR | $0.100 | $0.024 |
| Gemma 4 31B ITDevPass (LLM Gateway) | $0.100 | $0.024 |
| MiMo-V2-FlashHugging Face | $0.100 | $0.024 |
| Ministral 3 3BAmazon Bedrock | $0.100 | $0.024 |
| Nemotron 3 Ultra 550B A55Brouting.run | $0.100 | $0.024 |
| Ornith 1.5 35B A3BRunInfra | $0.100 | $0.024 |
| Qwen3.8 27BCortecs | $0.100 | $0.024 |
| Step 3.5 FlashHugging Face | $0.100 | $0.024 |
| Step 3.5 Flash 2603NanoGPT | $0.100 | $0.024 |
| Qwen3 Coder NextDevPass (LLM Gateway) | $0.108 | $0.026 |
| Qwen3.8 FlashDeep Infra | $0.113 | $0.027 |
| Qwen PlusAlibaba (China) | $0.115 | $0.027 |
| Qwen3.5 122B-A10BOrcaRouter | $0.115 | $0.027 |
| Qwen3.5 PlusMerge Gateway | $0.115 | $0.027 |
| Seed CharacterVolcengine Ark | $0.119 | $0.028 |
| Seed 1.6Ofox | $0.120 | $0.028 |
| Seed 1.6 VisionOfox | $0.120 | $0.028 |
| Seed 1.8Ofox | $0.120 | $0.028 |
| Gemini 3.1 Flash LiteKilo Gateway | $0.125 | $0.030 |
| Gemini 3.1 Flash Lite PreviewKilo Gateway | $0.125 | $0.030 |
| DeepSeek V4 Flash 0423Venice AI | $0.138 | $0.033 |
| DeepSeek V4.1 FlashAMD | $0.140 | $0.033 |
| GPT-5.2 CodexQiHang | $0.140 | $0.033 |
| Llama 4 Maverick 17B InstructAbacus | $0.140 | $0.033 |
| MiMo-V2-OmniXiaomi | $0.140 | $0.033 |
| MiMo-V2.5OpenRouter | $0.140 | $0.033 |
| Qwen3-VL PlusAlibaba (China) | $0.143 | $0.034 |
| Qwen3 Coder FlashAlibaba (China) | $0.144 | $0.034 |
| DeepSeek ChatOrcaRouter | $0.147 | $0.035 |
| DeepSeek ReasonerOrcaRouter | $0.147 | $0.035 |
| Gemini 3.5 Flash LiteKilo Gateway | $0.150 | $0.035 |
| Ministral 3 8BAmazon Bedrock | $0.150 | $0.035 |
| Mistral Small (latest)Eden AI | $0.150 | $0.035 |
| Mistral Small 4Eden AI | $0.150 | $0.035 |
| Qwen3.8 Flash NextAMD | $0.150 | $0.035 |
| Qwen3.6 FlashMerge Gateway | $0.165 | $0.039 |
| Qwen3.5 397B-A17BAlibaba (China) | $0.172 | $0.041 |
| GPT-5.4 nanoPoe | $0.180 | $0.043 |
| Llama 4 Scout 17B InstructDevPass (LLM Gateway) | $0.180 | $0.043 |
| Step 3.7 FlashStepFun (China) | $0.185 | $0.044 |
| Gemini 3.5 FlashUnoRouter | $0.186 | $0.044 |
| GPT-5.5UnoRouter | $0.188 | $0.044 |
| Grok 4.1 FastOfox | $0.200 | $0.047 |
| Grok 4.1 Fast (Reasoning)FrogBot | $0.200 | $0.047 |
| MAI-Code-1.1-FlashGitHub Copilot | $0.200 | $0.047 |
| Ministral 14BDevPass (LLM Gateway) | $0.200 | $0.047 |
| Ministral 3 14BDigitalOcean | $0.200 | $0.047 |
| Nemotron 3 Nano Omni 30B A3B ReasoningDeep Infra | $0.200 | $0.047 |
| Qwen3-Coder 480B-A35B Instructsubmodel | $0.200 | $0.047 |
| GPT-5.1 Codex miniPoe | $0.220 | $0.052 |
| MiniMax-M3EmpirioLabs AI | $0.225 | $0.053 |
| Gemini Flash-Lite LatestVertex | $0.250 | $0.059 |
| GPT-5.2QiHang | $0.250 | $0.059 |
| Trinity Large ThinkingArcee | $0.250 | $0.059 |
| Kimi K2.6routing.run | $0.275 | $0.065 |
| Kimi K2.7 Coderouting.run | $0.275 | $0.065 |
| Qwen3.6 PlusMerge Gateway | $0.276 | $0.065 |
| Qwen3.7 PlusAIHubMix | $0.282 | $0.067 |
| Codestral (latest)Eden AI | $0.300 | $0.071 |
| GLM-5.2CrofAI | $0.300 | $0.071 |
| Kimi K2.5NanoGPT | $0.300 | $0.071 |
| LongCat-2.0OpenRouter | $0.300 | $0.071 |
| Nova 2 LiteOpenRouter | $0.300 | $0.071 |
| DeepSeek V4 Prorouting.run | $0.348 | $0.082 |
| DeepSeek V4 Pro 0813CrofAI | $0.350 | $0.083 |
| Seed 2.1 TurboOfox | $0.354 | $0.084 |
| GPT-4.1 miniPoe | $0.360 | $0.085 |
| Qwen3 MaxOfox | $0.360 | $0.085 |
| Gemini 3.6 FlashKilo Gateway | $0.375 | $0.089 |
| GPT-5.4 miniXpersona | $0.375 | $0.089 |
| Devstral 2Eden AI | $0.400 | $0.094 |
| Devstral 2 (latest)Eden AI | $0.400 | $0.094 |
| GLM-5.3CrofAI | $0.400 | $0.094 |
| MiMo-V2.5-ProCrofAI | $0.400 | $0.094 |
| Mistral Medium (latest)Merge Gateway | $0.400 | $0.094 |
| Seed 2.0 CodeEmpirioLabs AI | $0.400 | $0.094 |
| Claude Opus 4.8UnoRouter | $0.425 | $0.100 |
| MiMo-V2-ProXiaomi | $0.435 | $0.103 |
| Inkling SmallOpenRouter | $0.450 | $0.106 |
| Kimi K2 ThinkingVercel AI Gateway | $0.470 | $0.111 |
| Seed 2.0 ProVolcengine Ark | $0.475 | $0.112 |
| Gemini Flash LatestOrcaRouter | $0.500 | $0.118 |
| Mistral Large (latest)Merge Gateway | $0.500 | $0.118 |
| Mistral Large 3Eden AI | $0.500 | $0.118 |
| Gemini 3 Pro PreviewQiHang | $0.570 | $0.135 |
| Qwen3 Coder PlusMerge Gateway | $0.574 | $0.136 |
| Palmyra X5Amazon Bedrock | $0.600 | $0.142 |
| Qwen3.6 27BPioneer | $0.600 | $0.142 |
| Hy4 previewVancine | $0.670 | $0.158 |
| Seed 2.1 ProOfox | $0.707 | $0.167 |
| Gemini 3.7 Flash302.AI | $0.750 | $0.177 |
| Gemini 3.8 Flash302.AI | $0.750 | $0.177 |
| GPT-5.4Xpersona | $0.750 | $0.177 |
| MAI-Code-1-FlashGitHub Copilot | $0.750 | $0.177 |
| Nova ProAmazon Bedrock | $0.800 | $0.189 |
| Qwen3.7 MaxMerge Gateway | $0.825 | $0.195 |
| Gemini 2.5 ProPoe | $0.870 | $0.206 |
| Seed EvolvingOfox | $0.884 | $0.209 |
| Claude Sonnet 4.6Xpersona | $0.900 | $0.213 |
| InklingDeep Infra | $0.950 | $0.224 |
| Sakana NamazuKilo Gateway | $0.950 | $0.224 |
| Gemini 3.1 Pro PreviewKilo Gateway | $1.00 | $0.236 |
| Grok Build 0.1AIHubMix | $1.00 | $0.236 |
| Qwen3.6 Max PreviewKilo Gateway | $1.03 | $0.243 |
| Vision SmallVispark | $1.05 | $0.248 |
| GPT-5OpenCode Zen | $1.07 | $0.253 |
| GPT-5-CodexOpenCode Zen | $1.07 | $0.253 |
| GPT-5.1OpenCode Zen | $1.07 | $0.253 |
| GPT-5.1 CodexOpenCode Zen | $1.07 | $0.253 |
| GPT-5.1 Codex MaxPoe | $1.10 | $0.260 |
| GPT-5 Chat (latest)Jiekou.AI | $1.13 | $0.266 |
| Kimi K2 Thinking TurboZenMux | $1.15 | $0.272 |
| Grok 4.20 (Non-Reasoning)Vertex | $1.25 | $0.295 |
| Grok 4.20 (Reasoning)Vertex | $1.25 | $0.295 |
| Grok 4.3302.AI | $1.25 | $0.295 |
| Muse Spark 1.1EmpirioLabs AI | $1.25 | $0.295 |
| Muse Spark 1.2Abacus | $1.25 | $0.295 |
| Muse Spark 1.3EmpirioLabs AI | $1.25 | $0.295 |
| MiMo-V2.5-Pro-UltraSpeedXiaomi | $1.30 | $0.308 |
| Claude Sonnet 5UnoRouter | $1.44 | $0.340 |
| GPT-5.6 TerraXpersona | $1.50 | $0.354 |
| GPT-5.3 CodexPoe | $1.60 | $0.378 |
| Qwen3.8 MaxVancine | $1.60 | $0.378 |
| DeepSeek V4 Pro 0423Merge Gateway | $1.65 | $0.390 |
| Qwen3.8 Max 0902Ofox | $1.71 | $0.404 |
| Mistral Medium 3.5evroc | $1.73 | $0.407 |
| GPT-4.1Poe | $1.80 | $0.425 |
| Kimi K2.7 Code HighspeedAIHubMix | $1.90 | $0.449 |
| Gemini 3.1 Pro Preview Custom ToolsAIHubMix | $2.00 | $0.472 |
| GPT-5.6 SolCloudflare AI Gateway | $2.00 | $0.472 |
| Grok 4.5AIHubMix | $2.00 | $0.472 |
| Grok 4.6Vertex | $2.00 | $0.472 |
| Kimi K3CrofAI | $2.00 | $0.472 |
| Qwen3.8 2.4T A95BOpenRouter | $2.00 | $0.472 |
| Qwen3.8 Max PreviewCharm Hyper | $2.00 | $0.472 |
| Command AEden AI | $2.50 | $0.591 |
| Command A ReasoningCohere | $2.50 | $0.591 |
| Claude Fable 5Xpersona | $3.00 | $0.709 |
| Vision MediumVispark | $4.21 | $0.994 |
| Claude Opus 4.6Poe | $4.30 | $1.02 |
| Claude Opus 4.7Poe | $4.30 | $1.02 |
| Claude Opus 5302.AI | $5.00 | $1.18 |
| Fugu UltraRequesty | $5.00 | $1.18 |
| GPT-5.5 InstantZenMux | $5.00 | $1.18 |
| Vision LargeVispark | $7.37 | $1.74 |
| Claude Fable 5.1302.AI | $10.00 | $2.36 |
| Claude Mythos 5Azure | $10.00 | $2.36 |
| GPT-6 Astra302.AI | $10.00 | $2.36 |
| GPT-5 ProJiekou.AI | $13.50 | $3.19 |
| GPT-5.2 ProPoe | $19.00 | $4.49 |
| GPT-6 Astra (Fast)Vercel AI Gateway | $20.00 | $4.72 |
| GPT-5.5 ProPoe | $27.27 | $6.44 |
| GPT-5.4 ProAzure | $30.00 | $7.09 |
Input only: what the model charges to read the context once. Output is billed separately and depends on what you ask for. Each row carries the cheapest listed provider for that model, which is often a gateway rather than the lab that trained it.
Why the window is not the target
Two things move when you send more context.
The first is the bill, and it is exactly linear. Input is priced per million tokens, so doubling the context doubles the cost every time. A model having room to spare is not a discount.
The second is harder to price. Retrieval accuracy does not stay flat across a full window. Liu et al. found that performance "is often highest when relevant information occurs at the beginning or end of the input context, and significantly degrades" in the middle, "even for explicitly long-context models". Chroma's sweep of 18 models across 8 input lengths reported that "across all experiments, model performance consistently degrades with increasing input length". Neither result says long context is useless. Both say the last half million tokens are not free even when you can afford them.
What a repository actually weighs
Across 60 public repositories tokenized on 2026-09-11, the median assembles into 236,218 tokens. Only 38% fit a 128,000 token window, 43% fit 200,000 and 72% fit 1,000,000. The spread runs from 34,860 to 3,044,662 between the 10th and 90th percentiles.
So for most projects the question is not whether a model can hold the codebase. It is whether sending all of it is the cheapest way to get the answer.
Filtering does less of that work than people expect. On a fresh clone the
defaults remove a median of 22.7% of tokens, and more than half of that is test
files dropped by name; a clone has no node_modules to remove in the first
place. What moves the number past that is deciding that a whole directory does
not belong in this prompt.
Cross-link
How many tokens is a codebase has the measurement behind the 236,218 figure and the distribution around it. Token estimation covers the tokenizer that produces the count on the result screen, and Token costs covers the output settings that change how much of a window a bundle consumes. File filtering is where you decide what goes in. ChatGPT Projects source limit is the other kind of ceiling, a file count per Project, which is why a folder goes in as one bundle.
References
- Lost in the Middle: How Language Models Use Long Contexts, Liu et al., TACL vol. 12, 2024, DOI 10.1162/tacl_a_00638
- Context Rot: How Increasing Input Tokens Impacts LLM Performance, Chroma Research, 2025
- models.dev, the catalogue this page prices against