# Open-weight models with the longest context windows

The longest-context open-weight models on the catalog reach 1M tokens: Llama Guard 4 12B leads at $0.20 per million input tokens. A 1M-token window holds roughly 750,000 words, an entire policy library or codebase in one prompt, on weights you can fine-tune and own.

## Ranked list

1. **Llama Guard 4 12B**. Meta · 1M context · $0.20 in / $0.20 out per M tokens · Llama community. https://allocate.network/models/meta-llama-guard-4-12b
2. **Llama 4 Scout Instruct (17Bx16E)**. Meta · 1M context · $0.18 in / $0.59 out per M tokens · Llama community. https://allocate.network/models/meta-llama-4-scout-17b-16e-instruct
3. **Ministral 3 14B Instruct 2512**. Mistralai · 256K context · $0.20 in / $0.20 out per M tokens · Apache 2.0. https://allocate.network/models/mistral-ministral-3-14b-instruct-2512
4. **Qwen3.5 9B FP8**. Qwen · 256K context · $0.17 in / $0.25 out per M tokens · Not listed. https://allocate.network/models/qwen-qwen3-5-9b
5. **Qwen3 235B A22B Instruct 2507 FP8 Throughput**. Qwen · 256K context · $0.20 in / $0.60 out per M tokens · Apache 2.0. https://allocate.network/models/qwen-qwen3-235b-a22b-instruct-2507-tput
6. **Qwen3-VL-8B-Instruct**. Qwen · 256K context · $0.18 in / $0.68 out per M tokens · Apache 2.0. https://allocate.network/models/qwen-qwen3-vl-8b-instruct
7. **Pearl-ai Gemma-4-31B-it-pearl**. pearl.ai · 256K context · $0.28 in / $0.86 out per M tokens · Not listed. https://allocate.network/models/pearl-gemma-4-31b-it
8. **Gemma 4 31B-it FP8**. Google · 256K context · $0.39 in / $0.97 out per M tokens · Apache 2.0. https://allocate.network/models/google-gemma-4-31b-it
9. **Qwen3 Next 80B A3b Instruct**. Qwen · 256K context · $0.15 in / $1.50 out per M tokens · Apache 2.0. https://allocate.network/models/qwen-qwen3-next-80b-a3b-instruct
10. **Qwen3 Next 80B A3b Thinking**. Qwen · 256K context · $0.15 in / $1.50 out per M tokens · Apache 2.0. https://allocate.network/models/qwen-qwen3-next-80b-a3b-thinking

## Data table

| Rank | Model | Input | Output | Page |
| --- | --- | --- | --- | --- |
| 1 | Llama Guard 4 12B | $0.20 | $0.20 | https://allocate.network/models/meta-llama-guard-4-12b.md |
| 2 | Llama 4 Scout Instruct (17Bx16E) | $0.18 | $0.59 | https://allocate.network/models/meta-llama-4-scout-17b-16e-instruct.md |
| 3 | Ministral 3 14B Instruct 2512 | $0.20 | $0.20 | https://allocate.network/models/mistral-ministral-3-14b-instruct-2512.md |
| 4 | Qwen3.5 9B FP8 | $0.17 | $0.25 | https://allocate.network/models/qwen-qwen3-5-9b.md |
| 5 | Qwen3 235B A22B Instruct 2507 FP8 Throughput | $0.20 | $0.60 | https://allocate.network/models/qwen-qwen3-235b-a22b-instruct-2507-tput.md |
| 6 | Qwen3-VL-8B-Instruct | $0.18 | $0.68 | https://allocate.network/models/qwen-qwen3-vl-8b-instruct.md |
| 7 | Pearl-ai Gemma-4-31B-it-pearl | $0.28 | $0.86 | https://allocate.network/models/pearl-gemma-4-31b-it.md |
| 8 | Gemma 4 31B-it FP8 | $0.39 | $0.97 | https://allocate.network/models/google-gemma-4-31b-it.md |
| 9 | Qwen3 Next 80B A3b Instruct | $0.15 | $1.50 | https://allocate.network/models/qwen-qwen3-next-80b-a3b-instruct.md |
| 10 | Qwen3 Next 80B A3b Thinking | $0.15 | $1.50 | https://allocate.network/models/qwen-qwen3-next-80b-a3b-thinking.md |

Provider list prices from the Allocate catalog, checked 2026-07-08. Billed price is list plus the 7% transaction fee.

## What long context is worth

Long context replaces retrieval plumbing for bounded corpora: instead of chunking and fetching fragments, the model reads the whole source. The tradeoff is per-request cost, which grows with the tokens actually processed, and memory on the serving side, where the KV cache grows linearly with context.

Open weights change the economics of long-document fine-tuning too: training examples that are whole documents need a window that holds them, which this list ranks directly.

## Common questions

### What is the longest context window on an open model?

1M tokens, on Llama Guard 4 12B, at $0.20 per million input tokens at list.

### Does long context cost more?

Per token, no: you pay the same list price per million tokens. Per request, yes, because you send more tokens. A full 1M-token prompt on a $0.18 model costs about $0.18 at list before caching.

## Related

- [What is a context window?](https://allocate.network/glossary/context-window)
- [What is the KV cache?](https://allocate.network/glossary/kv-cache)
- [GPU VRAM calculator](https://allocate.network/tools/gpu-vram-calculator)

---

[HTML page](https://allocate.network/best/open-models-with-the-longest-context) · [Machine-readable catalog](https://allocate.network/catalog.json)