A context window of around one million tokens holds a mid-sized codebase, a stack of contracts or months of chat logs in one request. Several models on GetVibe API offer a window that size, from flagships to low-cost flash models. They all run on one OpenAI-compatible endpoint and one key.
The list
Context and output limits come from our catalog (caps.contextWindow in https://getvibeapi.com/api/catalog). Prices are live.
| Model | Context | Max output | Images | JSON mode | Price |
|---|---|---|---|---|---|
| Claude Opus 5.5 | 1,000,000 | 128,000 | yes | yes | $0.1067 input · $0.5334 output per 1M tokens |
| Claude Sonnet 5 | 1,000,000 | 64,000 | yes | yes | $0.3334 input · $1.6667 output per 1M tokens |
| GPT-5.6 Sol | 1,050,000 | 128,000 | yes | yes | $0.0667 input · $0.40 output per 1M tokens |
| Gemini 3.1 Pro | 1,000,000 | 65,000 | yes | yes | $0.2515 input · $1.5086 output per 1M tokens |
| Gemini 3.8 Flash | 1,000,000 | 65,000 | yes | no | $0.0097 input · $0.0483 output per 1M tokens |
| DeepSeek V4.1 Flash | 1,000,000 | 393,216 | yes | yes | $0.0112 input · $0.0445 output per 1M tokens |
| GLM-5.3 | 1,048,000 | 128,000 | no | no | $0.10 input · $0.3143 output per 1M tokens |
| GLM-5.3 Flash | 1,048,000 | 128,000 | yes | no | $0.0112 input · $0.0371 output per 1M tokens |
Other 1M-class models are in the catalog too, including older Claude Opus versions and Gemini 3.7 Flash. The Models page shows the full list and lets you sort it.
Declared versus measured
Most numbers above are the window the model maker declares. For Claude Opus 5.5 we also measured it before publishing: a 732,379-token request was answered with details from both the start and the end of the prompt, and a request over 1,000,000 tokens was rejected at the model's limit. For the other models, test with your own data if you plan to run close to the limit.
How to use a long window well
- Put the question after the material. Send the documents first and the instruction last, so the model reads the task with everything already in context.
- Mark the boundaries. Wrap each file or document in clear delimiters with its name. It helps the model cite where an answer came from.
- Check both ends. When you evaluate a model for long inputs, ask about a detail near the start and one near the end of the same prompt.
- Leave room for the answer. Input and output share the window, so a prompt that fills it completely leaves no space for the reply.
Cost: every input token is billed
A full 1M-token request is billed for every input token, so long context is the most expensive way to call a model. Two things keep it in check:
- Prompt caching. If you ask several questions about the same material, keep it as an unchanged prefix. Cached input is billed at a fraction of the input rate. See the prompt caching guide.
- Pick the tier by task. For search-and-extract over a large input, a flash model with a 1M window may be enough. Save the flagships for reasoning across the material.
from openai import OpenAI
client = OpenAI(base_url="https://getvibeapi.com/v1", api_key="sk-pv-...")
corpus = open("all_docs.txt").read()
resp = client.chat.completions.create(
model="deepseek-v4.1-flash",
messages=[
{"role": "system", "content": "Answer only from the documents provided."},
{"role": "user", "content": corpus + "\n\nQuestion: which contracts mention a termination fee?"},
],
)
print(resp.choices[0].message.content)Next steps
Read the Claude Opus 5.5 1M context article, compare models on the Models page, or start with the documentation.