Find open-weight models with long context windows.
Bring the published context window and output limit together before you choose a model for long documents or a repository. This shortlist uses explicit open-weights evidence and a stated context threshold, with links back to the catalog sources.
174 language models have an open-weights flag and at least 128,000 listed context tokens. Up to 12 appear below, ordered by listed context. This is a capacity shortlist, not a retrieval-quality ranking.
Catalog snapshot · 12 rows shownTreat the window as a capacity limit
The threshold here is 128,000 listed context tokens. Context is the model’s reported capacity; maximum output is a separate limit. Your prompt, attached material, conversation and generated answer all need room within the provider’s actual serving limits.
A larger window does not prove better retrieval or reasoning across that window. Check a representative document, including facts near its beginning, middle and end. Keep the task and answer checks the same when you try another model.
Separate weights, license and hosting
An open-weights flag does not establish a permissive commercial license, free inference or a downloadable endpoint configuration. Open the model card to check its license and access conditions. Then check the provider route or the hardware you intend to use.
Quantization and serving settings can affect memory needs and model behavior. The catalog does not infer those settings from the model name. Tool calling and structured-output flags below identify reported capabilities; an absent flag means not reported, rather than a failed capability test.
How this view is assembled
Selection: language models explicitly flagged as open weights, with a finite listed context window of at least 128,000 tokens. At most 12 models, sorted by context and then name. This threshold is an editorial shortlist rule, not a benchmark. The catalog update date is not a license verification or evaluation date.
Long-context open-weight models
Model catalog specifications, with the original catalog sources
| Model | Context window | Maximum output | Capabilities to check | Sources and dates |
|---|---|---|---|---|
| Llama-4-Scout-17B-16E-Instruct | 1.3M tokens | 16.4K tokens | Tool calling listedStructured output listedLicense and hosting cost require a separate check | OpenRouter ↗Catalog updated Hugging Face ↗Catalog updated |
| Xiaomi: MiMo-V2.5 | 1.1M tokens | 65.5K tokens | Tool calling listedStructured output listedLicense and hosting cost require a separate check | OpenRouter ↗Catalog updated Hugging Face ↗Catalog updated |
| Xiaomi: MiMo-V2.5-Pro | 1.1M tokens | 235.9K tokens | Tool calling listedStructured output listedLicense and hosting cost require a separate check | OpenRouter ↗Catalog updated Hugging Face ↗Catalog updated |
| Xiaomi: MiMo-V2.6-Flash | 1.1M tokens | 131.1K tokens | Tool calling listedStructured output listedLicense and hosting cost require a separate check | OpenRouter ↗Catalog updated Hugging Face ↗Catalog updated |
| Xiaomi: MiMo-V2.6-Pro | 1.1M tokens | 131.1K tokens | Tool calling listedStructured output listedLicense and hosting cost require a separate check | OpenRouter ↗Catalog updated Hugging Face ↗Catalog updated |
| Meituan: LongCat 2.0 | 1M tokens | 262.1K tokens | Tool calling listedStructured output not reportedLicense and hosting cost require a separate check | OpenRouter ↗Catalog updated |
| DeepSeek-V4-Flash | 1M tokens | 943.7K tokens | Tool calling listedStructured output listedLicense and hosting cost require a separate check | OpenRouter ↗Catalog updated Hugging Face ↗Catalog updated |
| DeepSeek-V4-Flash-0731 | 1M tokens | 943.7K tokens | Tool calling listedStructured output listedLicense and hosting cost require a separate check | OpenRouter ↗Catalog updated Hugging Face ↗Catalog updated |
| DeepSeek-V4-Flash-Vision-Exp | 1M tokens | 262.1K tokens | Tool calling listedStructured output listedLicense and hosting cost require a separate check | OpenRouter ↗Catalog updated Hugging Face ↗Catalog updated |
| DeepSeek-V4-Pro | 1M tokens | 393.2K tokens | Tool calling listedStructured output listedLicense and hosting cost require a separate check | OpenRouter ↗Catalog updated Hugging Face ↗Catalog updated |
| DeepSeek-V4-Pro-0813 | 1M tokens | 393.2K tokens | Tool calling listedStructured output listedLicense and hosting cost require a separate check | OpenRouter ↗Catalog updated Hugging Face ↗Catalog updated |
| DeepSeek-V4.1-Flash-1017 | 1M tokens | 393.2K tokens | Tool calling listedStructured output listedLicense and hosting cost require a separate check | OpenRouter ↗Catalog updated Hugging Face ↗Catalog updated |
Where these rows come from
We join catalog specifications, recorded provider quotes and published evaluation records where exact model identities are available. Some observations come through aggregators or benchmark collections. Multiple listings do not establish independent confirmation of a claim.
Open the source beside a row to check the original evidence. A collection or check date is separate from a model release or evaluation date. Unknown values stay unknown. Recorded prices are in USD with their original billing units and settings.
See data coverage and sources →Compare for your own use case
Keep your task, model identity and provider configuration together. Explore the catalog yourself, or connect your AI to read the same source-backed records.